Home

A Visualization Toolkit for Large Geospatial Image Datasets Zachary

1. param curVtxAttrsBean Attributes of the currently selected vtx x return a String containing the URL for a properly formatted request to the DataLinkServlet for the resource given by lt code gt branch lt code gt at nav vertex lt code gt strNavVtxName lt code gt since JDK1 2 public static String formatViewResourceRequest Branch branch String strNavVtxName lol AttributeMapBean cur VtxAttrsBean x x x x x Returns a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt param strNavVtxName Nav vertex of requested resource param strDataVtxName The name of a data vertex for the type of resource requested param attrMap A set of name value pairings that define the attributes of the selected resource can be null return a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt since JDK1 2 public static String formatViewResourceRequest String strN
2. 88 5 2 1 View Requests ad wo Se ee ee Kom ag 89 5 2 2 View Resource Requests 08050048 90 5 2 3 Specifying Abstract State Values 90 5 3 How the DataLink Server Handles View Requests 91 5 3 1 How the Navigation Layer Handles a Select Vertex Request 92 5 3 2 How the Data Layer Fetches the Branches for a Data Type 9 5 4 How the DataLink Server Handles View Resource Requests 98 8 6 The Geospatial Coordinate Transformation Library 101 6 1 Problems With Existing Packages 102 6 2 GCTL Makes Reference Frames Explicit 103 6 2 1 Design Principles of GCTL 02 0 0 105 6 3 GCTL Understands Five Types of Coordinates 106 6 3 1 Datums are Models of the Earth s Surface 2 2 107 DL ST GOL sanea aes a a es ts ea Be es ee Fire 111 6 41 GOTL Data Type xo es Su au Aria eee ewe EE eS 111 6 4 2 Defining Local Grid Coordinate Systems 112 6 4 3 GCTL Can Convert Between Geodetic and Geocentric Coordi DALOS e vnc at Be Bees He Oe ee ee eos a es Sw Meee hs 113 6 4 4 GCTL Can Do Datum Shifts 115 6 4 5 GCTL Can Make Maps With Cartographic Projections 119 6 5 Chapter summary 22 2 20 K sa DE 2 u nenn 120 7 Advanced Features of the Extended Web Interface 121 7 1 Enhanced Visualizations 0000 ee eens 121 7 1 1 Extensions to the Map Viewer 122
3. 4 formatViewResourceRequestURL 5 openResourceStream All of these methods are essentially the same in their operation in that they are used to make requests to the DataLink Server given a branch or a set of attributes and possibly a history Some other methods implemented by the DataLinkDispatcher which may be useful for content handlers are summarized in Table 4 5 The formatViewRequest and formatResourceRequest methods return a string containing a GET protocol request such as the one from Section 4 2 These methods are useful for creating request URLs for making hyperlinks to branches or passing data to embedded applets formatViewRequestURL and formatViewResourceRequestURL do the same thing only they output their results in a Java URL class openResource Stream is capable of making the same requests as the above methods except that it opens an InputStream directly to the resource which can be used to open files for reading and parsing There are several different versions of each of these request formatting methods each of which takes a different combination of arguments Some versions operate directly on instances of the Branch class and data from the curPathBean Others sl enable the programmer to use the attribute map data structures from Section 4 3 2 to make requests using an arbitrary set of attribute bindings To go into the details of each of these variations would be overly fastidious However the inter
4. GCTL performs computations only on LLA coordinates expressed in radians This is usually not the format that the source data is in or the one that the programmer In keeping with the conventions of Section 6 2 1 it would seem like a good idea to define a Vector2D data type for cartographic projections I considered this but decided to stick with Vector3D because it facilitates matters when cartographic projections need to be thought of as planar projections in three space One real example of this is when the City group extrudes building facades from a campus basemap in cartographic coordinates to check against the 3D geometry of the City Data images Some alternate representations of altitude are based on isoclines in the earth s gravitational field EUROCONTROL and IfEN 1998 111 wants to use in the output so the LLA class has two methods for switching back and forth between radians and degrees Each of these outputs a new LLA object with the value of the original expressed in the alternate angle format 1 LLA toDeg const 2 LLA toRad const 6 4 2 Defining Local Grid Coordinate Systems ECEF coordinates are the common language which GCTL uses to connect most of the coordinate systems it can work with together Every coordinate system ob ject in GCTL except for the cartographic projections has methods for converting to and from ECEF coordinates or WGS84 geocentric coordinates which are the same thing There
5. try listen for connections until we are told to shutdown while m_bShutdown 174 m_lLastWakeupTime System currentTimeMillis listen for new connection socket null try blocks for 5s then checks for shutdown signal socket m_serverSocket accept catch InterruptedIOException e continue accept timed out connection new Connection socket synchronized m_vConnections m_vConnections add connection tryt connection out write HI WELCOME TO m_strName toUpperCase n THE CURRENT NUMER OF CONNECTIONS IS m_vConnections size n getBytes catch Exception e shutdown synchronized m_vConnections if m_vConnections null for int n 0 n lt m_vConnections size n connection Connection m_vConnections get n try connection out write SERVER SHUTTING DOWN n getBytes connection out flush catch Exception e disconnect connection m_vConnections null if m_serverSocket null m_serverSocket close catch I0Exception e 175 what can you do write to System err end class OutputStreamServer D 2 MonitorServlet java import java util import java 10 import javax servlet import javax servlet http import util monitor x x x This server page is used to invoke another JSP or servlet with standard out redirected to an Ou
6. vanishing points check box beneath the image windows the node viewer displays a family of lines radiating from each vanishing point To help the user find the source of the van ishing points they are marked with stars in the sphere viewport window Figure 7 8 shows the vanishing points of a node visualized using a node viewer 132 Figure 7 9 The Map Viewer s baselines visualization The Baselines Visualization The other global features that are extracted by post processing are baselines Baseline directions in the City dataset closely match the directions between adjacent nodes The Map Viewer s visualization of baselines is meant to show how accurate the cor respondence between baseline directions and the actual directions between the nodes is The Map Viewer draws baselines in a fashion that is analogous to the way it draws vanishing points except that center point of a baseline line segment is located at the midpoint between two adjacent nodes As it does for vanishing points the Map Viewer matches the length of a baseline segment based to the distance between the two adjacent nodes which share it Figure 7 9 shows the Map Viewer s baselines visualization This visualization is turned on using a check box labeled baselines which is located underneath the map image The node viewers allow the user to see the baseline directions from a node s per spective When the baselines visualization for a node viewer i
7. COS_67P5 rAlt rDistZ rCosLat rRad else rAlt v3 2 rSinLat rRad m_rEccSq 1 0 if bAtPole rLat atan2 rSinLat rCosLat return LLA rLat rLon rAlt end Ellipsoid XYZtoLLA 167 168 Appendix D Source Code For the OutputStreamServer and MonitorServlet D 1 OutputStreamServer java package util monitor import java io import java net import java util Debugging JSPs and servlets can be a real pain because they execute on the remote server within the servlet engine Even if the servlet engine elects to output any System out printlin messages they probably just get printed to a console windoow on the server which is no good to somebody logged in on a development machine I wrote this simple server class to get around that issue It allows clients to login to a specified port using telnet or a similar client interface to receive data sent to the OutputStreamServer OutputStreamServer extends OutputStream which makes it possible to redirect standard out or standard error to an QutputStreamServer This lets a developer insert System out println debug statements in his her servlet code x xx r KF KF HF KF KF F 169 and see them printed in a console window at a development machine author ZMB public class QutputStreamServer extends OutputStream implements Runnable members private int m_nPort private long m_lLastWakeupTime private ServerSocket m_
8. View resource requests include an extra parameter which tells the DataLink Server that it is ok to resolve the attribute bindings directly to a resource This makes requests for physical data which can often occur in large numbers quite a bit more efficient It is a good practice to use view resource requests whenever possible in 3If you remember Figure 2 1 from the chapter on the architecture of the DataLink Server this invokes the data path labeled Path 2 82 Current Path all nodes fixed Dataset all nodes Node 67 translatedfhalt rotated initial Mransiated Ne mi mosaiced 0000 cam May 15 2002 6 00 PM 0004 cam May 15 2002 6 00 PM Figure 4 1 Screen shot of a navigation context menu content handlers It is always possible to determine whether a branch corresponds to a resource by calling the Branch class isResource method One caveat that is important to keep in mind however is that view resource requests made with abstract state values instead of literal state values will evaluate to garbage because it is necessary to invoke the data matching process in order to resolve abstract states to literal states This is is only a concern when making requests to the DataLink Server for data that does not correspond to one of the branches in the curPathBean because all state attributes in these branches are completely specified with both abstract and literal states 4 5 Making Menus With t
9. epipolar geometry node locations as well as many other features of the data Finally it provides a remapping of the physical data structure of the dataset to a simpler logical hierarchy more suitable for web browsing The Pre existing Visualization Tools At the core of the pre existing web interface are the visualization applets This section describes the functionality of each of these applets briefly The Map Viewer The Map Viewer shows the positions of the nodes superimposed upon a map of campus The node locations are shown in different colors corresponding to their level of post processing refinement The Map Viewer can zoom in to show a 5See Scaleable Absolute Position Recovery for Omni Directional Image Networks Antone and Teller 2001 for more information about the translation georeferencing stage 18 Node s Position and Nearest Neighbors ale EG Key mosaiced rotated translated Show C baselines I vanishing points V adjacencies view full scale map The lines befween nodes on fhe mini map connect each node fo one of ifs nearest neighbors Click on one of these lines fo view the epipolar geometry of fhe two neighboring nodes Figure 1 1 A screen shot of the Map Viewer zoomed in to show node 104 close up view of a region of campus and can display adjacencies as lines which connect node locations Clicking on a node location redirects the user to a CGI generated HTML document that displays
10. ing that all of the branches have earlier timestamps than the cache file This is a quick way of checking whether any of the source files have been modified or whether new ones have been added but it will not be able to detect changes in the names of symbolic links or whether source files have been deleted To detect these changes the caching content handler computes a checksum from the size t of each of the branches and their names s according to this formula N S gt t mod hashcode s 7 6 i 0 This checksum is compared against the checksum of the cached file and if they match the cached file is served Otherwise it is regenerated and then served The combined use of the branch name and size captures all of the possible changes that occur to the data It is not foolproof though Some changes may not be detected if there is a hash code collision for two different files of the same size or if both the new name and the old name of a renamed file generate hash codes that are less than the file size However since most changes to the data involve adding a group of nodes or re numbering all the nodes this hash code strategy should detect most changes in conjunction with the time stamps check This caching strategy has been employed for the node locations data as well as the baselines data used by the Map Viewer on the browse page of the extended web interface It dramatically speeds up the operation of this part of the w
11. language that is used to define them 2 2 2 The Core of the DataLink Server is the DataLinkServlet At the heart of the DataLink Server is the DataLinkServlet Java servlet which provides the navigation data location infrastructure needed to drive a web interface based on the specification captured in the model file The DataLinkServlet is backed by two components the DataModel and the NavModel which are software representa tions of the data and navigation models from the model definition file The DataModel and the NavModel are the modules of the DataLink Server which implement the data and navigation layers respectively A block diagram of this architecture is shown in Figure 2 3 The detailed inner workings of the DataLinkServlet the DataModel and the NavModel are related in Chapter 5 2 2 3 The Front End Consists of Content Handlers As mentioned earlier the presentation layer is controlled by content handlers which may be Java server pages servlets or HTML documents for static content Content 39 Presentation Layer Content Handlers JSPs Serviets HTML Applets EGR DataLinkServlet Navigation Layer NavModel Data Layer DataModel Figure 2 3 Architecture of the DataLink Server handlers give the web designer freedom to customize the look and feel of the web interface and flexibility in controlling the way the data is presented to the user A content handler could simply present links to the raw data mu
12. www w3 org 2001 XMLSchema instance xsi no NamespaceSchemaLocation http city lcs mit edu 8080 data2 models mode 1s xsd gt lt data model block gt lt dataModel gt lt This URI specifies where the dataset lives gt lt home href gt lt This is where the data vertex type definitions should go gt lt This is where the data vertex instances are created gt lt dataModel gt lt navigation model block gt lt navModel gt lt Here we define the navigation vertices gt lt navModel gt lt modelDefs gt This code listing shows the where the two principal blocks of the model file belong The modelDefs element is the root element of the model file Contained within it are the dataModel and navModel elements which encapsulate the data model and navigation model definitions the term element refers to a construct of the form lt element gt element data lt element gt 47 Use GET request URLs Restricted From navigation vertex names and at tribute names GET request URLs navigation vertex names and at tribute names delimits history elements in navigation vertex names DataLink Server requests Character Table 3 1 Reserved characters 3 2 1 An XML Schema is Used For Validation In the previous example the modelDefs element references an XML schema file lt http city lcs mit edu 8080 data2 models models xsd gt which encodes the defi
13. amp l a 2 LLA NAD83toNAD27 const LLA amp lla In order to use these methods one must create an instance of the NADConv class initialized with the grid for the geographic region of interest This is done using the NADConv loadGrid function load the grid file for continental US if nadConv loadGrid CONT_US fprintf stderr Couldn t load continental US grid file s nadConv getLastErr As the example shows GCTL has a defined moniker for each of the grid files These are are listed in Table 6 3 The example also demonstrates the use of the NADConv class error reporting mechanism the function getLastErr which pro duces a string containing a description of the last error encountered Errors can arise either in loading the grid file or when converting coordinates if the converted coordi nate lies outside of the specified grid region When this happens the value returned by NAD83toNAD27 or NAD27toNAD83 will be HUGE_VAL HUGE_VAL h where HUGE_VAL is the value of infinity defined in the standard C math library 118 Alaska AK Continental US CONT_US Florida FL Hawaii HI Pruerto Rico Virgin Islands PR_VI St George Is AK ST_GRG St Lawrence Is AK ST_LRNC St Paul Is AK ST_PAUL Tenessee TN Washington Oregon WA_OR Wisconsin WI Table 6 3 GCTL NADCON Grids 6 4 5 GCTL Can Make Maps With Cartographic Projections GCTL also has the capability to perform cartographic projections It doe
14. that can decode radiance encoded rgb images This both eliminates the need to generate jpg versions of the images for web viewing and the problems with color saturation see Section 7 1 4 for more information about the RGBImage class Figure 1 2 is a screen shot of the Mosaic Viewer displaying some of the enhancements described in this thesis see Section 7 1 The Epiploar Viewer The Epipolar Viewer is a two panel version of the Mosaic Viewer Each panel shows the view from one node of a pair of adjacent nodes In each view the neighboring node is marked by a blue cross The Epipolar Viewer allows the user to project a ray from the center of one view and see the corresponding line as viewed from the neighboring node For a pair of perfectly aligned nodes the ray should appear to intersect the same points in space in both views thus allowing the user to visually inspect the quality of the position and orientation estimates generated by the post processing routines The user can select using a drop down menu which set of pose data to use when rendering the epipolar lines This allows the user to observe how the position and orientation estimates improve with each post processing stage Figure 1 3 shows a screen shot of the Epipolar Viewer being used to examine two views of The Great Sail In Figure 1 3 the Epipolar Viewer has been augmented with some of the advanced features described in this thesis see Section 7 1 6Calder Alexander The Gr
15. their measurements are the cause of non linearities in shifts between the older datums and newer ones As Figure 6 1 shows these non linearities between the datums can be quite significant In reality no one ever performs the shift between these two particular datums because they are almost identical There is generally a much larger difference between other pairs of datums however S This data was obtained from the National Geodetic Survey National Geodetic Survey 2002al 116 Latitude 126 W 116 W 106 W 96 W 86 W 76 W 66 W 29 N 24 NI ae 2 26W 116W 106W 96 W rer wks Figure 6 1 NAD83 NAD27 datum shifts in arc seconds 117 This problem was of particular interest to the City group because we wanted to unify the City dataset with a base map of campus that is encoded in US State Plane coordinates which are a cartographic projection in the NAD27 datum Fortunately the National Geodetic Survey has done some work in establishing some grids of very precise control points in both the NAD27 and NAD83 Datums Accurate datum conversions between NAD27 and NAD83 for individual points can be obtained by linearly interpolating between nearby grid points in what is known as the NADCON method National Geodetic Survey 2002a GCTL has a class which implements this conversion called NADConv This class has two methods for converting between NAD28 and NAD83 geodetic coordinates 1 LLA NAD27toNAD83 const LLA
16. tion their coordinate axes relative to each other Thus performing a datum shift on geocentric coordinates is essentially a change of basis which can be done by applying a homogenous matrix such as Eq 6 1 to the geocentric coordinate vectors GCTL s datum classes have two methods for doing this 1 Vector3D toWGS84 const Vector3D amp v3 const 2 Vector3D fromWGS84 const Vector3D amp v3 const Thus to change from NAD83 to WGS84 using an instance of the NAD83_Datum class NAD83Datum one would apply function 1 like this Vector3D v3ECEF NAD83Datum toWGS84 v3NAD83 To transform between two arbitrary datums using GCTL one first converts geo centric coordinates from the first datum to WGS84 Then to get coordinates in the second datum one transforms the intermediate results from WGS84 to the second datum To reduce computation by doing this all in one step one can compose the to WGS84 matrix of the first datum with the from WGS84 change of basis matrix of the second These matrices are defined in GCTL using a Mat44 Matrix 4 x 4 class which has functions for matrix multiplication operations so it is easy to do this Grid Shifts Are Used to Deal With Older Datums Many datums such as the North American Datum of 1927 NAD27 were surveyed using conventional techniques before the advent of satellites and GPS This poses a problem because many of these datums are still in use today but the inaccuracies in
17. while at the same time being easily maintainable by a web architect Folding several logical organizations of the data into one allows a user to explore the full multi dimensionality of the data space Such a system should give the user the capability to navigate to the same piece of data though several alternate paths to accommodate for the fact that various users will have different perceptions about what the logical arrangement of the data should be and it should only provide navigation options for data that is physically present The DataLink Server takes several important steps towards realizing this goal It provides the web developer with centralized easy to modify configuration mecha nism an XML document describing the physical data and the set of valid navigation paths to be made available to users It also supports a modular front end interface constructed from JSPs Java servlets and HIML documents called content handlers each of which handles the presentation of data at a particular set of locations within the navigation hierarchy The DataLink Server has been successfully deployed as the backbone of the City Scanning Project Dataset where it facilitates the navigation of the more than 11 000 geospatial images the City group has collected since the spring 29 of 2000 and the data we have extracted from them 2 1 The DataLink Server Maps Virtual Navigation Hierarchies to Real Data The DataLink Server implements a three level
18. 7 1 2 Extensions to the Node Viewers 124 7 1 3 New Global Visualizations 130 7 1 4 The New Web Interface Has Support For Radiance Encoded Images 6 04 os owt se baw eRe woe eG ES g 134 7 2 Content Handlers Can Perform Data Caching 136 7 3 User Auditing and Debugging 2 000 4 138 8 Contributions and Concluding Remarks 141 A The DataLinkDispatcher API 145 B A Complete Listing of the Model File Schema models xsd 159 9 C An Implementation of Toms Method For Converting Between Geo centric and Geodetic Coordinates 165 D Source Code For the OutputStreamServer and MonitorServlet 169 D 1 OutputStreamServer java 2 2 Cm onen 169 D 2 MonitorServlet java 2 22 Coon 176 10 List of Figures 1 1 1 2 2 1 2 2 2 3 2 4 3 1 3 2 3 3 3 4 o 1 6 1 A screen shot of the Map Viewer zoomed in to show node 104 19 A screen shot of the Mosaic Viewer enhanced with some of the exten sions described in this thesis 2 2 moon 21 A screen shot of the Epipolar Viewer comparing the geometry of two views of The Great Sail 2 2 22mm nn 22 Interactions between the abstraction layers of the DataLink Server 31 A simple set of models for a classical music library 39 Architecture of the DataLink Server 2 2 2 2 2 m mn 40 How the content handler for the composer vertex might format the data for a Composer 2 2 Hmm e e 41 A screen shot
19. DataLink Server 55 Default values of navigation vertex parameters 65 NavPathBean routines for use in content handlers 77 Map data structures for holding attribute sets 78 BranchClass routines for use in content handlers 80 Branch methods for content handlers 80 Other useful methods provided by the DataLinkDispatcher 81 GCTL Datum Classes 2 2 22 2 oo EEE 108 GOTL Ellipsoid Classes 4 amp 55 22 244 Ha 2 0 0 44444 109 GCTL NADCON Grids zu 2 4 ara a ae eK a en amp 119 13 14 Chapter 1 Introduction and Background Information 1 1 Background Information The goal of the City Scanning Project is to develop an end to end system for automat ically acquiring 3D CAD models of urban environments under uncontrolled lighting conditions i e outside under natural light Teller et al 2001 To achieve this goal the City Group has constructed a sensor for automatically acquiring pose augmented imagery image data coupled with position and orientation estimates and a suite of post processing algorithms to refine the position estimates and extract intrinsic parameters about building features Using these tools the City Group has collected over 500 sets of pose augmented images which make up we believe to be the largest dataset of geospatial imagery in existence The author had previously developed a world wide web based interface that allows people from
20. File 46 3 2 1 An XML Schema is Used For Validation 2 48 3 2 2 Reserved Characters 0 0 000 2 eee eee 48 3 3 Creating a Data Model 64 ss Lau 2 2 4G el 48 3 3 1 Data Vertex Type Definitions 0 49 3 3 2 Declaring Data Vertex Instances 55 3 4 Writing a Model To Direct Navigation 58 3 4 1 Defining Navigation Vertices 59 3 0 Chapter SUMmMary ua Name eh a oe eh OS ee a OA 71 Content Handlers 73 4 1 Content Handlers Are JSPs Servlets or HTML Documents 73 4 2 The Format of DataLinkServlet Requests 74 4 2 1 The history Parameter 79 4 3 Data Structures for Content Handlers 76 4 3 1 The DataLinkServlet Communicates to Content Handlers Us ing the curPathBean 0 0200004 76 4 3 2 Data Structures for Attributes 77 4 3 3 Data Structures for Branches 4 79 4 4 The DataLinkDispatcher Is Used to Format Requests to the DataLink DONE a Be ee de Gee ayes Met ae te Spans oe Sea Ar hae 79 4 4 1 View Requests vs View Resource Requests 82 4 5 Making Menus With the BranchHistoryScripter 83 46 Chapter Summary aa e ac woe a an Se Oe ee a Re Ewe De x 86 Operation of the DataLink Server 87 5 1 Data Structures for Curbing Computational Complexity 87 5 2 DataLink Server Requests Come in Two Flavors
21. Node Viewers The Mosaic Viewer and Epipolar Viewer have been significantly upgraded The most noticeable change to the user interface is the addition of the sphere viewport This is a smaller window beneath the planar projection window which shows the unwrapped spherical texture A screen shot of the Mosaic Viewer with the sphere viewport provided in Figure 7 3 The sphere viewport allows the user to get some global context about where the mosaic viewer is looking by displaying a view of the entire sphere adjacent to the perspective from the node As Figure 7 3 shows the direction of the current view is indicated by a small gold target in the sphere viewport This viewpoint target is This a cylindrical projection that has been split lengthwise and unwrapped to make a flat 2D image 124 Figure 7 3 The sphere viewport window interactive The user can drag and drop it to a new location in the viewport window and the view from the node perspective window will change accordingly This makes locating objects that are visible in the unwrapped texture much easier and improves the overall experience of using the Mosaic Viewer and the Epipolar Viewer The Node Viewers Have a Heads Up Display The node viewers are more than just fancy image viewers They are also equipped to make use of the pose data that accompanies each node in the City dataset One thing that the node viewers can do with this information is provide the user with a hea
22. StatePlaneProjector spProj 2001 LLA lla new LLA 65 73 0 104 USSP sp spProj project LLA printf x f y f sp 0 spl1 The odious Proj 4 code has metamorphosed into into something in which the programmer always knows what type of coordinates are in use and what type of arguments are appropriate for each transformation For a more provocative example investigate the following piece of code which is used to transform a point in Cartesian coordinates expressed in the NAD83 datum to a point in State Plane coordinates lla NAD83Datum XYZtoLLA v3NAD83 lla nadConv NAD83toNAD27 11a sp spProj project lla Without knowing something about geospatial coordinate systems as you may not just yet it s hard to understand the purpose of each of these steps For example you wouldn t know that State Plane coordinates are expressed relative to the NAD27 datum whatever that is so before the State Plane projection can be applied to the coordinates they must be shifted from the NAD83 datum in which they are currently expressed to NAD27 What is clear in this example however is that in the first step some Cartesian coordinate expressed relative to some frame of reference named NAD83 is being transformed to an LLA coordinate also expressed relative to NAD83 Then in step two the reference frame of the LLA coordinate is shifted from NAD83 to NAD27 Finally the coordinate is projected to state plane
23. abstract state of the data so in this case it will make a request with only the abstract state of the state attribute defined In other cases such as when the client wants to make a request for only the trans_set4 data it may make a request with the literal state of the refinement attribute specified Table 3 2 lists some examples using the refinement attribute which illustrate each of the possible value pairings a state attribute may assume Data Vertex Type Signatures Define the Mapping of Attributes and Data Data vertex type signatures define the mapping between attributes and physical data and conversely tell the DataLink Server how to derive attribute values from an instance of a data type in the dataset This example vertex from the City Scanning Project Dataset should help to illustrate how signatures work 93 lt vertexType name image gt lt attributes gt lt int name imgId gt lt attributes gt lt signature gt 4d rgb imgId lt signature gt lt vertexType gt This sample is the data vertex type definition for an image in the City dataset Each node in the City dataset consists of a collection of numbered rgb images with filenames of the format 0000 rgb 0001 rgb etc The four digit number in the name of each image is the image s ID which is captured by the imgId attribute of the image data vertex type The signature that maps the four digit image ID in the image s filename to the imgId attribute is si
24. as specified in the navigation model getNavVtxName The name of the navigation vertex corresponding to this branch class getBranches A Branchl of all of the branches in this class isResource true if the branches are resources leaf vertices false if they are navigation vertices Table 4 3 BranchClass routines for use in content handlers Method Return Value getBranchName The formatted name of this branch as specified in the navigation model getAttributes An AttributeMapBean containing the attribute bindings of this branch getLastModified The timestamp of the branch in milliseconds if the branch is for a navigation vertex that is associated with a data type 1 if the source vertex is purely navigational getSize If this branch is associated with a physical datum this method returns its size in bytes it returns 0 if the asso ciated navigation vertex is purely navigational Table 4 4 Branch methods for content handlers 80 Method Return Value getDataLinkServer Returns a string containing the URL to the DataLink Server Formats the navigation history into a path like string delimited by characters formatHistoryPath String pHistory sortBranches Branch Sorts a branch array lexicographically by name pBranches Table 4 5 Other useful methods provided by the DataLinkDispatcher 1 formatViewRequest 2 formatViewResourceRequest 3 formatViewRequestURL
25. as the Earth and is also concentric with it The NAD27 datum on the other hand is slightly smaller than the Earth is and is localized at a point such that it is tangential to the Earth s surface at Mead s Ranch in Kansas Consequently it is only accurate in North America Hofman Wellenhof et al 1997 Geodedic and geocentric coordinates are always defined with respect to a datum You may previously thought that latitude and longitude coordinates were universal but in actuality since they are dependent on the Earth s shape which is approximated by using some reference ellipsoid they are correlated with a datum Geocentric coordinates are the same way They are defined with respect to a Cartesian coordinate system originating from the center of a reference ellipsoid which can be defined precisely mathematically not with respect to the Earth s true center which is only known approximately For this reason it is not possible to transform geodetic or geocentric coordinates to any other coordinate system unless the datum in which they are expressed is known This is the reason why GCTL contains constructs such as lla NAD83Datum XYZtoLLA v3NAD83 This statement converts geocentric coordinates to geodetic coordinates where both coordinate systems are expressed relative to the NAD83 Datum GCTL also contains mechanisms for performing shifts between datums The datums which GCTL knows about are listed in Table 6 1 Datums usually
26. e A teleological naming convention that makes mastering the techniques for ma nipulating geospatial coordinates facile I have also contributed the following web visualizations for exploring the City Scanning Project Dataset online e A Map Viewer which superimposes the refined location estimates of the nodes in the City dataset for the first time ever on a CAD map surveyed using conven tional techniques that demonstrates the precision of the City Scanning Group s 3D reconstruction algorithms e Visualizations of edge and point features vanishing points and baselines in the City dataset e An Epipolar Viewer for comparing the quality of registration of stereo geospatial image pairs It is my hope that the visualizations I have created will prove to be an informative educational tool for other researchers in the field of machine vision or curious web surfers who would like an opportunity to learn more about the City Scanning Project These tools should also turn out to be an effective way for members of the City Scanning team itself to validate and evaluate the image processing and machine vision 142 algorithms being developed as part of the City Scanning Project Lastly I believe that the DataLink Server and the Geospacial Coordinate Transformation Library themselves will provide powerful utilities for the software development community at MIT and beyond to create web portals to large datasets and work with geospatial coo
27. edges are dis played as purple lines superimposed on the viewing window which move along with the image as the viewpoint is rotated or zoomed The intersections visualization is turned on by a check box labeled intersections and draws the point features as pur ple targets Side by side screen shots of these visualizations are shown in Figure 7 5 In the second view both the edge and intersection visualizations have been activated for clarity 126 Figure 7 5 Edge and Point Feature Display The Epipolar Visualization Has Been Improved The epipolar line drawing algorithm of the Epipolar Viewer has been meliorated for the new visualization suite The old epipolar line algorithm projected a nearby point and a distant point on the same epipolar line and connected the two projected points to draw the epipolar line in the node s perspective window The new algorithm begins at a point 1 meter distant from the source node and interpolates the epipolar line by incrementally plotting points farther and farther away from the node s eye As the epipolar line heads off towards an infinitely distant vanishing point one meter increments do not make much of a difference AX AY in the projected points so the epipolar line algorithm increases the step size logarithmically by doubling it every time 4 1 20 yi Yo lt 25 pixels where Zo yo are the projected coordinates of the previously drawn point and 21 y are the coor
28. files The camera files from previous stages are kept in a directory named after the post processing stage that generated them Each stage can be described briefly as follows Feature Detection During this stage edge and point features are extracted from each of the raw images for use in later post processing stages This stage outputs edge and point feature files Mosaic At this stage the overlapping regions of the raw images are aligned The images are then blended together and resampled to form a spherical texture The camera files generated by this stage for a given node have improved orientation estimates relative to each other This stage outputs spherical textures and mosaiced camera files Rotation Edge features are used in the rotation phase to detect vanishing points points where families of parallel lines seem to converge when viewed from the optical center of the node The rotation process relies on the assumption that nearby nodes will share some common vanishing points because they view some of the same building facades The nodes are rotated in an attempt to align the shared vanishing points of adjacent nodes The result is further refinement of the orientation estimates This stage outputs vanishing points node adjacencies and rotated camera files Please see Spherical Mosaics with Quaternions and Dense Correlation Coorg and Teller 2000 for more information about the algorithms used in the mosaic stage 3
29. gt lt vertexType gt lt biography file gt lt vertexType name bio_doc gt lt signature gt s txt composerName lt signature gt 96 lt vertexType gt lt picture of composer gt lt vertexType name protrait_img gt lt signature gt s jpg composerName lt signature gt lt vertexType gt lt This is where the data vertex declarations will go gt lt dataModel gt This data model says that our music library is contained within a directory with some name like classical which should be stored in the datasetName at tribute Inside this directory are some subdirectories such as Johannes Brahms Modest Mussorgsky etc which are named after the composers in the library In each composer directory there are two subdirectories with fixed names audio_files and bio Each bio directory contains a biography file and a portrait of the com poser and held within each audio_files directory are the pieces written by com poser whose filename is constructed from the composer s name followed by the title of the work and the opus number i e Johannes Brahms Sonata for Clarinet and Piano no 1 op 120 mp3 Notice that attributes are declared at the highest level of the hierarchy in which they appear and declarations for them are omitted from vertex type definitions for data that lie beneath this node in the data hierarchy This is because when instances of these data vertex types are connected
30. gt s s 4s firstName middleName surname lt signature gt lt signature gt s 4s firstName surname lt signature gt lt vertexType gt lt dataModel gt 69 lt navModel gt lt vertex name composer type composer_dir gt lt child name opus number gt lt child name pieces gt lt child name portrait gt lt child name bio gt lt branchName gt s s surname firstName lt branchName gt lt branchClassName gt Composers lt branchClassName gt lt vertex gt lt navModel gt lt modelDefs gt Now the heading for composer branches will be Composers and each composer branch itself will have the form Brahms Johannes Notice that to do this some new attributes firstName middleName and surname had to be created and the signature for the compser_dir had to be modified to extract the values for these attributes In this listing I ve only shown the changes relevant to this discussion on name formatting but clearly other vertices in the navigation and data models would have to be augmented with the new attributes as well The main thing to take away from this example however is how the two branch formatting commands branchName and branchClassName are used Assigning Content Handlers to Navigation Vertices To assign a custom content handler to a navigation vertex the contentHandler com mand is used Here is an example of this command s usage lt vertex name compos
31. hash function to place the attribute sets in a hashed set is a quick and dirty linear time way of checking for equivalent attribute sets 5 4 How the DataLink Server Handles View Re source Requests View resource requests are considerably simpler for the DataLink Server to process The influence of the navigation layer is minimally invasive because the view resource request explicitly specifies a data vertex and there are no children to explore The navigation layer simply requests that the data layer resolve the physical path to the data using the supplied attributes The data layer can do this without invoking the matching and binding process described in Section 5 3 2 because the implicit assumption of a view resource request is that the resource is completely specified by the supplied attributes If for some reason when trying to evaluate the physical path the data layer finds that a necessary attribute was not supplied it reports an error and returns nothing In fact aside from retrieving the ancestor chain for the request data vertex the extent of the data model s involvement corresponds to carrying out the same process it uses for accumulating the base path Steps 1 and 1 1 of the matching algorithm except that instead of terminating at the head vertex it evaluates its signature as well For clarity here is an explicit enumeration of these steps 1 Initialize the base path to be the root path of the dataset 2 Loop over
32. have acquired all of the knowledge necessary to begin designing the architecture of powerful web gateways to large datasets 3 1 The Model Definitions are Described in XML The modeling constructs used to control the DataLink Server are written in XML XML was chosen as the basis of the modeling platform in lieu of some custom language The definitive resource on XML is Extensible Markup Language XML The World Wide Web Consortium 2002a lt http www w3 org XML gt 49 and file format because it has a number highly desirable qualities that make it ideal for this application One property in particular that makes XML a good candidate for the job of defining the models is that its structure consisting of elements that are localized within the context of other parent elements closely parallels the directed graph structure of the models themselves In fact when an XML document is parsed the data structure that is produced is a tree Other features of XML which make it an excellent choice for this task are Powerful XML parser distributions are available in Java and C e Error checking and reporting for an XML file is automatic when it is coupled with a validating parser and a document type specification such as an XML schema or a DTD e The use of a markup language makes the models file easy to read and compre hend e The syntax of a model file written in XML will be familiar and intuitive to any developer w
33. in a pair of attribute sets must be compared but the hashing step can still be used to help if a new hashing function is defined for attribute sets The hash code for an attribute set A containing attribute bindings Qo Q an that is used in this implementation of the DataLink Server is defined to be t N Ha gt Heine name a string values a 5 2 i Here the function values a returns the string value of a in the case of state attributes values a returns the abstract string value Hstring is the same hash func tion as that used for branch merging Eq 5 1 The hash function for attribute sets is designed so that if two sets contain the same key value pairs they will always have the same hash code regardless of the order of the attributes in the sets In this case 3Recall from Section 3 3 1 that the preferential ordering of literal states is given by the sequence in which their signatures appear in the state attribute s definition in the model file 4The hash code of the empty attribute set is defined to be 0 97 equivalent but not necessary equal attribute sets will always have the same hash code as well because the hash code used for state attributes is the hash code of their abstract state This is exactly the behavior that is needed in this case because it will always trigger a comparison when attribute sets are equivalent but avoid need less comparisons most of the the times when they are not Using this
34. in urban environments GPS World pages 20 30 Brown et al 2000 Brown S Burdick R Falkner J Galbraith B Johnson R Kim L Kochmer C Kristmundsson T and Li S 2000 Professional JSP WROS Press Inc Birmingham United Kingdom second edition Calder 1966 Calder A 1966 The Great Sail Massachusetts Institute of Tech nology Cambridge Coorg and Teller 2000 Coorg S and Teller S 2000 Spherical Mosaics with Quaternions and Dense Correlation International Journal of Computer Vision 37 3 259 273 181 De Couto 1998 De Couto D S J 1998 Instrumentation for Rapidly Acquiring Pose Imagery Master of engineering thesis MIT EUROCONTROL and IfEN 1998 EUROCONTROL and IfEN 1998 WGS84 Im plementation Manual Manual University FAF Munich Germany Evenden 1995a Evenden G I 1995a Cartographic Projection Procedures for the UNIX Environment A User s Manual Technical report remote sensing org 26 July 2001 lt ftp ftp remotesensing org pub proj OF90 284 pdf gt 2 7 MB Evenden 1995b Evenden G I 1995b Cartographic Projection Procedures Re lease 4 Interim Report Technical report remote sensing org 26 July 2001 lt ftp ftp remotesensing org pub proj PROJ 4 3 pdf gt 9 7 MB Evenden 1995c Evenden G I 1995c Cartographic Projection Procedures Re lease 4 Second Interim Report Technical report remote sensing org 26 July 2001 lt
35. layer then looks for every instance of data vertex type 7 in the data model and examines its ancestry to construct a chain of data vertices from each matching data vertex to the a root vertex in the data hierachy The set of chains revealed by this process describes all of possible paths to data of type T The data model then returns a set of complete attribute bindings for all the paths in this set which are consistent with the attribute values specified by the navigation layer For example suppose our music library contained only four pieces the Brahms sonatas for clarinet and piano no 1 and no 2 op 120 Beethoven s symphony no 2 op 36 and Vivaldi s The Four Seasons op 8 Selecting the branch of navigation vertex composer for Brahms from the root vertex would define the current set of attribute bindings to be composer Brahms datasetName classical The navi gation layer would then examine each of the child vertices of composer to determine 4There is no restriction on the number of root vertices that may exist in the data model 37 the next set of navigation options that should be made available to the user To dis cover the possible options for vertex pieces the navigation layer would request the set of all branches for data vertex type mp3 for which composer Brahms from the data layer The data layer would then look in the data model for instances of vertex type mp3 and find only one PIECES The chain for t
36. minimum and maximum radiance values in the image and Ce is the 0 255 log radiance value of the pixel to be re mapped When images encoded this way are displayed as is they appear to be grayed out or solarized In order to correctly render the images the log radiance pixel values must be re mapped to a linear 0 255 scale which is done using this formula R Bir Fini en a 7 5 255 Rinreshold Rmaz gt Rmin The Rinreshold Value is used to set a saturation threshold for the pixels Pixels with radiance values greater than this will be assigned a value of 255 for the corresponding channel If the results of Formula 7 1 4 are negative the pixel is assigned a 0 value for channel c The key to making log radiance encoded images look good on a computer display is to find the right value for Rynreshold Ihe rgb image class has a simple heuristic for doing this It computes the cumulative distribution function for the radiance values in the image for all channels and sets the threshold to the radiance value at which the CDF is 90 This means that when the image is displayed 90 of the pixels will be unsaturated This is a pretty good heuristic for most cases because most images only have a few really bright pixels from the sky while the rest of the image falls in a more limited dynamic range All of the applets in the City Scanning Dataset web interface which display images are also equipped with slider bars that the user can manipulate to adj
37. name 2 gt lt signature gt state 2 signature 1 lt signature gt lt signature gt state 2 signature 2 lt signature gt lt state gt lt stateVar gt 50 A state attribute can assume any one of a discrete set of abstract states The abstract states are defined by lt state gt elements inserted within the body of the state attribute definition Every abstract state declaration must include at least one lt signature gt A signature is a character pattern that the data link server uses to identify items in the physical data A signature is a pattern which matches a filename or part of a filename in the case of state attributes These state signatures are basically keywords used by the data layer to determine the abstract state of state attributes from properties of the physical data If the value of a state attribute extracted from a filename matches a particular signature of abstract state a then the data layer knows that the abstract state of the attribute is a To see how this works let s examine an example from the City dataset Recall from Section 1 1 1 that after a City Scanning node is acquired it enters a pipeline of post processing stages Each of these stages improves the position and orientation information associated with each image in the node and outputs a new refined set of camera files for those images The camera files for each post processing stage are stored in a directory named after that stage such as initial mo
38. nd where n is the number of steps and d is the step size t is the interpolated point on the parametrized 128 epipolar line where the interpolation is between two points t and ta with correspond ing z values z and 22 If t is chosen such that the distance from the eye at this point is 0 then is the distance from the eye at the interpolated point The advantage of this approach is that the step size is always uniform in screen space which saves computation where the previous algorithm would have to double the step size and try again when ever the screen step becomes too small The disadvantage however is that screen space interpolation will generally not result in integer distance markings One additional modification to the Epipolar Viewer that would fruitful however is to compute the vanishing point of the epipolar line in order to accurately plot its ultimate endpoint The vanishing point can never be reached by interpolation since it is infinitely far away so the epipolar lines plotted by the current algorithm terminate somewhat prematurely though still at points much further than any objects in the images The vanishing point of an epipolar line is easy to compute however so it would not be difficult to modify the existing Epipolar Viewer to terminate epipolar lines at their vanishing points The epipolar line from the perspective of the source camera is given by the points along 7 rt yt t This line project
39. of the content handler for the node vertex of the City dataset au ce ix amp Gi ae a Oe Se ee we re a te 42 Data model of a classical music library 55 Navigation model of the classical music library 59 Listing of the classical lib vertex aaaea 61 Another listing of the classical lib vertex 62 Screen shot of a navigation context menu 83 Data paths through the DataLink Server 92 NAD83 NAD27 datum shifts in arc seconds 117 The new Map Viewer 2 2 LK on nn 123 7 2 Screen shot of the Map Viewer zoomed in on the Green Building nodes 124 7 3 The sphere viewport window 2 08 125 7 4 The heads up display 0 0 2 0080004 126 7 0 Edge and Point Feature Display 0 00 020 127 7 6 Anepipolar viewing of Building 18 with distance markings on the epipoles128 7 7 Vanishing points visualization of the nodes along Mass Ave 131 7 8 The node viewers vanishing point visualization 132 7 9 The Map Viewer s baselines visualization 133 7 10 The node viewers baselines visualization 134 7 11 The MonitorServlet 2 2 om nme 139 12 List of Tables 3 1 3 2 3 3 3 4 4 1 4 2 4 3 4 4 4 5 6 1 6 2 6 3 Reserved characters 0 a a a a 48 Valid state pairings listed by example 53 String format expression tokens used by the
40. of the information it needs directly from the curPathBean In addition to the curPathBean the program mer only has to provide the method with a writer to the output stream of the request response that is being issued by the content handler The content handler writer object are available from lt http developer netscape com viewsource smith_menu smith_ menu html gt Javascript is a tricky thing to make compatible with all of the different web browsers that are out there which is why it is a good idea as well as a time saving practice to use something that is tried and true 6 This is JSP speak for the document that is output by the content handler JSP or servlet 80 can now insert the navigation context menus anywhere in the body of the document using the insertBranchHistory method lt BODY BGCOLOR lt strBGColor gt gt lt CENTER gt lt H2 gt The MIT City Scanning Project Dataset lt H2 gt lt CENTER gt lt Insert the navigation context menus here gt lt P gt lt BranchHistoryScripter insertBranchHistory new PrintWriter ou t curPathBean gt 4 6 Chapter Summary Having read this chapter the reader should now have acquired the knowledge nec essary to begin developing content handlers for a web interface created using the DataLink Server This provides a flexible and modular way of creating a useful and aesthetically pleasing front end to a web interface defined using the modeling language of Chapt
41. storing the branches themselves a BranchClass stores some data which is shared by all of the branches in the class This includes the name of the source navigation vertex and a flag which specifies whether the branches correspond to resources physical files Table 4 3 lists the methods of BranchClass which are of interest to authors of content handlers The branches themselves are represented by instances of the Branch class The methods this class provides for getting at the data are given in Table 4 4 In an effort to reduce redundancy the DataLinkServlet returns all of the attribute bind ings of the current navigation path which are shared by all of the branches in the curPathBean Only the attribute bindings which are specific to individual branches are stored in the branches themselves Thus the complete set of attributes for a given branch is a union of the current path s attributes and the attributes stored in the Branch class for that branch curPathBean getCurVtxAttrs U branch getAttributes 4 4 The DataLinkDispatcher Is Used to Format Re quests to the DataLink Server Now that the details of the data structures for storing attributes and branches have been articulated the use of the DataLinkDispatcher can be investigated The DataLinkDispatcher implements several different versions of these five methods for interfacing with the DataLinkServlet 19 Method Return Value getName The name of this branch class
42. to the data In addition any changes in 7One might consider running these scripts in a crawler like mode to automatically incorporate additions and deletions However this involves both the creation and removal of symbolic links in the public HTML directory of the web interface so it is likely to cause irritating 404 Not Found messages and other disruptions if the dataset is modified at the same time it is being browsed by a user due to the caching that is performed by most web browsers Also the DataLink Server is capable of incorporating updates in real time whereas the crawler scripts would have to be run episodically 23 the organization of the physical data or its layout on the web requires modification of the Perl scripts themselves which necessitates at a minimum decent working knowledge of the dataset the operation of the scripts and the Perl programming language The DataLink Server provides a more versatile solution to this problem The Data Link Server is a web application consisting of Java servlets and Java Server Pages JSPs that provide an abstraction function between a particular web layout and a physical data store The DataLink Server allows a web designer to define one or more alternative logical arrangements of the data using a simple XML based mod eling language The hierarchical organization of these logical arrangements may be completely different from the layout of the physical data Using the XML description
43. to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which must lt b gt not lt b gt be null since JDK1 0 public static String formatViewRequest String strNavVtxName SimpleNamedValueMap attrMap Returns a String containing the URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which must lt b gt not lt b gt be null This method is intended for use in applications where Java 1 2 may not be used param strNavVtxName Name of the nav vertex to view param attrMap A set of name value pairings that define the attributes of the selected vertex param pHistory An array containing the names of the n vertices visited thus far which pHistory 0 pointing to the first vertex the root and pHistory n pointing to the latest x x return a String containing the URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which must lt b gt not lt b gt be null x since JDK1 0 public static String formatViewRequest String strNavVtxName SimpleNamedValueMap attrMap 148 String pHistory I I III II III II III III formatViewRequestU
44. tools are packaged as software applications which offer the developer no way to access them from within pro gram code The closest many of these packages come is the ability to read coordinates from a text file or other source and output them in another for mat but there is no standard input format and the user s source data is usually not in the correct one to begin with 2 Those packages which exist in library form are difficult to use without a solid foundation in geodesy and geospatial coordinate transforma tion The problem with many of these platforms is that the coordinate frames of reference are implicit Therefore it is difficult for inexperienced user s to figure out exactly how to chain multiple transformations together There is such a large assortment of geospatial coordinate transformation packages and ap plications that it is too intractable to list them all here A few of the most useful ones are Proj 4 Warmerdam 2000 Evenden 1995a Evenden 1995b Evenden 1995c and NADCON National Geodetic Survey 2002a which is part of a large collection of free tools the Geodetic Toolkit National Geodetic Survey 2002b provided by the National Geodetic Survey Geodesy is the science of the Earth s shape which is not as trivial as you may think and models for expressing coordinates on its surface 102 3 Poor usage of non object oriented models is pervasive throughout these tools A lot of the t
45. unordered element groupings in which each element of the group may appear more than once This restriction is not terribly inconvenient as long as one remembers 49 it when writing model files In point of fact this rule carries throughout any XML document the elements must appear in a fixed order Attribute Definitions There are two components to every attribute construct a type declaration and a name In essence attributes are declared in the same way as variables are in any programming language Attribute names can consist of any character sequence that does not include the amp or characters In addition attribute names cannot be one of the three keywords which are used in requests to the DataLink Server view resource and history Integer Attributes True to their name integer attributes can hold any integer value Integer attribute definitions have the form lt int name attributeName gt String Attributes The value of a string attribute can be a character sequence of any length String attribute definitions look like lt string name attributeName gt State Attributes State attributes are a little more complicated than string and integer attributes State attribute definitions look like lt stateVar name attributeName gt lt abstract states gt lt state name abstract state name 1 gt lt signature gt state 1 signature lt signature gt lt state gt lt state name abstract state
46. user could take through the dataset is to begin by selecting the opus number of a piece to find This would take 2The data models discussed in this document will typically be models of directory trees This is a reflection of the first target application of the DataLink Server purveying the navigation infrastructure for the City Scanning Project Dataset which is stored in an elaborate directory structure It should be noted however that the data model is merely an abstraction and that although the software implementing this abstraction is tailored to modeling directory trees it would be easy to extend the data model to provide an interface to any hierarchical data storage system such as a database 39 the user to the opus number vertex Figure 2 2 does not completely specify what the behavior of the system should be at this point but there are two possibilities 1 The system could display link for every piece in the dataset with the selected opus number as well as a link for every composer in the dataset 2 The system could display only the the links to the composer branches and omit the links to the audio files because the attributes accumulated along the path traversed thus far are not enough to completely specify the path to any piece of music It is possible to configure navigation vertices with either behavior Before reading any further be sure to note that the edges in the data model and navigation model graphs point in opposi
47. value classical gt lt vertex gt Properties can be set for any navigation vertex and there can be as many proper ties as there are attributes Once properties are set they are inherited by all navigation vertices descending from the one in which the property was defined so for our exam ple the composer pieces portrait and bio vertices will only expand to branches in the classical library and not to their counterparts in the punk rock dataset Derived Properties are Used to Promote Attributes So far the navigation model we have scribed only describes the right hand portion of the diagram in Figure 3 2 where the user follows the edge from the classical_lib vertex to the composer vertex and down to the music image and biography files This next code segment adds the left hand path through the opus number vertex comments have been included to indicate the changes lt navModel gt lt vertex name classical lib gt lt child name composer gt lt vertex gt lt vertex name composer type composer_dir gt lt child name opus number gt lt added this edge gt lt child name pieces gt lt child name portrait gt lt child name bio gt lt vertex gt lt added this vertex gt lt vertex name opus number gt lt child name composer gt 63 lt child name pieces gt lt derivedProperty name opusNum source mp3 gt lt vertex gt lt vertex name pieces type mp
48. vanishing points IV adiacencies view full scale map The lines between nodes on the a a mini map connect each node to one of its nearest neighbors Click on one of these lines to view the epipolar geometry of the two neighboring nodes saturation threshold 4 Images The images you see on this page constitute one node from the City Scanning Project dataset The raw images seen at the saturation threshold bottom of this page share a common optical center but are rotated into various orientations that together tile a hemisphere During ae EEE the mosaic stage of post processing these images are more accurately aligned with each other and combined to form the spherical ne Wi aa Se texture seen above The spherical texture is better viewed using the Mosaic Viewer at the right which allows you to view the a EEE Basen node from its optical center and rotate the viewing angle along the horizontal and vertical axes I edges intersections The images in the City Scanning Project Dataset are all high dynamic range HDR images Although the images are encoded using the lossless SGI rgb format with color values ranging from 0 255 these values actually represent the log of the true radiance values The following equation captures the relationship between the rgb pixel values and the corresponding Status Figure 2 5 A screen shot of the content handler for the node vertex of the City dataset vanishing points The A
49. 1 4 Return the bindings in Voinaings 2 Base case Once the head vertex is reached obtain a listing of all of the data located at the base path 3 For each datum in the listing attempt to bind the null valued bindings of v to the attributes of the datum If successful create a new attribute set A by inserting the new bindings into A and add it to the vector of bindings Vbindings 96 4 If A contains any state attributes that are specified by abstract state only ex amine the A to see if any pair of them is equivalent If so unify the implicated pair 9 Return Vpindings Attribute Set Equivalence and Unification When the client makes a request using abstract state values for state attributes it doesn t care what the literal state of the data is as long as it is the most preferred state when there are more than one options available When the data layer uncovers two sets of bindings which are the same except for the literal states of a pair or pairs of state attributes the two sets of attribute bindings are equivalent The data layer should only return the set with the preferred literal state whose value should be anulled To do this the data layer must in Step 4 compare each of the A for equivalence This is reminiscent of the merging procedure discussed earlier in Section 5 3 1 In this case the problem is exacerbated by the fact that in order to check for equivalence each pair of corresponding attributes
50. 3 gt lt vertex name portait type portait_img gt lt vertex name bio type bio_doc gt lt root name classical lib gt lt navModel gt In this example the lt derivedPropery gt command has been used to promote the opusNum attribute to a higher level in the hierarchy The branches displayed by the DataLink Server for the opus number vertex when viewed from the classical lib vertex will be the set of all unique opus numbers that exist in the dataset The source argument of the derivedProperty tells the DataLink Server how to determine what the possible values of the derived property are In this case the DataLink Server is instructed to do this by examining all of the mp3 files that are consistent with the attribute values set by the user s current path through the navigation hierarchy When the user is at the classical lib vertex no attribute values have been accumulated so the branches displayed for the opus number vertex are the opus numbers for every mp3 file in the dataset When the user is at the composer vertex composerName has been bound to a value so the set of opus number branches includes only the opus numbers of works by the selected composer Attribute Redirection Attribute redirection can be used to reassign a named value to the name of an at tribute for a particular vertex by using a redirect command in the body of a navi gation vertex Ihe redirect command has the syntax lt redirect source name of s
51. A Visualization Toolkit for Large Geospatial Image Datasets by Zachary Modest Bodnar Submitted to the Department of Electrical Engineering and Computer Science in partial fulfillment of the requirements for the degree of Master of Engineering in Electrical Engineering and Computer Science at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY May 2002 Turns 22m Zachary Modest Bodnar MMII All rights reserved The author hereby grants to MIT permission to reproduce and distribute publicly paper and electronic copies of this thesis in whole or in part TTS INSTITUTE OF TECHNOLOGY Q Ro JUL 31 2002 LIBRARIES D aeee n nee PRIR TEEN Department of Electrical Engineering and Computer Science May 24 2002 Certified by er Seth Teller Associate Professor of Computer Science and Engineering nr A gt _ Thesis Supervisor Accepted Dy aa EAN Arthur C Smith Chairman Department Committee on Graduate Students A Visualization Toolkit for Large Geospatial Image Datasets by Zachary Modest Bodnar Submitted to the Department of Electrical Engineering and Computer Science on May 24 2002 in partial fulfillment of the requirements for the degree of Master of Engineering in Electrical Engineering and Computer Science Abstract This thesis describes a suite of tools that were used to build an internet gateway with powerful 3D and 2D visualizations and intuitive directed navigation to an expansive col
52. A node is adjacent to any of its 3 or sometimes 4 nearest neighbors and any other nodes within a certain radius Adjacencies are used to determine which nodes are likely to share common features such as vanishing points 4See Automatic Recovery of Relative Camera Rotations for Urban Scenes Antone and Teller 2000 for more information about the rotation stage 17 Translation and Georeferencing As a node is moved in a straight line points on the spherical texture seem to emerge from a common point as they move toward the viewer and converge at an antipodal point as they move away from the viewer These two points define a baseline By attempting to align baselines directions with the directions between two nearby nodes the translation process refines the relative position estimates of the nodes Then GPS estimates from the sensor are then used to scale translate and rotate the camera positions to obtain new estimates that best fit the surveying data This stage outputs baselines for each pair of adjacent nodes and camera files with revised position estimates 1 1 2 Description of the Existing Interface The existing web interface consists of a number of HTML documents CGI scripts and Java applets which allow the user to browse the nodes view the camera files and feature files as well as view thumbnails of the raw images It also includes a number of visualization tools which allow the user to view the spherical textures adjacencies
53. AttributeMapBean in bean form as instances of the AttributeBean class This class has a very simple interface which includes a getName and a getValue method State attributes are stored as a subclass of AttributeBean StateAttributeBean which has an additional method getState for retrieving the abstract state of a state attribute A Word About State Attribute Values All of the map classes for storing attributes can work with attribute values represented as strings Attributes values of any type can be represented as strings but there are some special considerations for state attribute values in string form Recall from Section 3 3 1 that state attributes can have an abstract state with no literal state or both an abstract state and a literal state When a state attribute has a literal state the string version of its value is simply the literal state it is in i e auto_rot When a state attribute has only an abstract state its string value is the name of the abstract state prefixed with a character as in translated This reflects the way state 18 attributes values are passed in a query string to the DataLink Server The getValue method of the StateAttributeBean class will output string representations of state attribute values according to this convention 4 3 3 Data Structures for Branches The DataLinkServlet bundles all of the branches for a single navigation vertex into a BranchClass class In addition to
54. PI for Writing Content Handlers The DataLink Server package provides two crucial classes for interfacing with the DataLinkServlet that facilitate writing content handlers The DataLinkDispatcher class implements about 20 static methods for formatting requests to the DataLinkSer vlet and processing the data returned by it These include methods for sorting branches opening streams to resource files via the DataLinkServlet and formatting URLs or hyperlinks that access DataLinkServlet given a set of attribute values which may be encoded in one of serveral different types of data structures The DataLink Server also includes a class for formatting special context navigation menus BranchHistoryScripter for use in content handlers The menus created by this class are activated by mousing over the name name of one of the branches visited along the user s current path through the navigation hierarchy which are listed at the top of the page or wherever the author of the content handler decides to place them and enumerate all of the other alternative branches the user could have chosen at that point in the path Selecting one of these options jumps to that branch In the sample screen in Figure 2 5 the user has activated the context menu for the node branch The complete documentation of the DataLinkDispatcher and the BranchHistoryScripter are located in the chapter devoted to content handler development Chapter 4 There are more than 500 n
55. RL IIIIIIIII III TTT TT Returns the URL for a propererly formated request to view the branch lt code gt branch lt code gt of nav vertex lt code gt strNavVtxName lt code gt The caller should ensure that the history of vertices that have been visited so far are passed to the method using the lt code gt pHistory lt code gt argument if the context requires that current navigation path be preserved The history can be obtained directly from the NavPathBean forwarded by the DataLinkServlet param branch Branch to view param strNavVtxName Nav vertex of the selected branch param curVtxAttrsBean Attributes of the currently selected vtx can be obtained from the NavPathBean forwarded by the DataLinkSerlvet param pHistory An array containing the names of the n vertices visited thus far which pHistory 0 pointing to the first vertex the root and pHistory n pointing to the latest x x x x return The URL for a view request for the supplied arguments since JDK1 2 public static URL formatViewRequestURL Branch branch String strNavVtxName AttributeMapBean curVtxAttrs Bean String pHistory Returns the URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which may be null param strNavVtxName N
56. Real rRad m_rMajorAxis sqrt 1 0 m_rEccSq rSinLat rSinLat Real rA rRad lla alt cos lla lat return Vector3D rA cos rLon rA sin rLon rSinLat rRad 1 0 m_rEccSq lla alt SISISSIS ST XYZtoLLA IIIIII Transforms a cartesian coordinate to geodedic lat lon alt coordinate in radians Both coordinates must be expressed relative to this ellipsoid LLA Ellipsoid XYZtoLLA const Vector3D amp v3 const Real rLon rLat rAlt bool bAtPole false if v3 0 0 0 rLon atan2 v3 1 v3 0 else 166 if v3 1 gt 0 rLon PI 2 0 else if v3 1 lt 0 rLon PI 2 0 else bAtPole true rLon 0 0 if v3 2 gt 0 0 north pole rLat PI 2 0 else if v3 2 lt 0 0 south pole rLat PI 2 0 else return LLA PI 2 0 rLon m_rMinorAxis Real rDistZsq v3 0 v3 0 v3 1 v3 1 Real rDistZ sqrt rDistZsq distance from Z axis Real rVO v3 2 AD_C Real rHO sqrt rVOxrVO rDistZsq Real rSinBO rVO rH0 BO is an estimate of Bowring aux variable Real rCosBO rDistZ rH0 Real rV1 v3 2 m_rMinorAxis m_rEccPrimeSq rSinBO rSinBO rSinBo Real rSum rDistZ m_rMajorAxis m_rEccSq rCosB0 rCosBO rCosB0 Real rHi sqrt rVi rVi rSum rSum Real rSinLat rVi rHi Real rCosLat rSum rHi Real rRad m_rMajorAxis sqrt 1 0 m_rEccSq rSinLat rSinLat if rCosLat gt COS_67P5 rAlt rDistZ rCosLat rRad else if rCosLat lt
57. Seth Teller Title Associate Professor of Computer Science and Engineering Acknowledgments Patience is the cardinal virtue of any good teacher and therefore I would first like to acknowledge my thesis advisor Prof Seth Teller who despite the impressive number of students projects and courses he supervises still finds time to sit down with any one of his students to share pointers about such humble subjects as debugging with gdb and working with a UNIX shell as eagerly as he provides insights on graphics algorithms mathematics and other hard core topics of computer science I must also thank Mom for her unfaltering support and encouragement Sis for her ever consummate advice my brother Eric for restraining himself from dissuading me from coming to MIT and Dad to whom I owe my love of the outdoors and my respect for the natural world I should thank Prof Patrick Henry Winston for reinvigorating in me a passion for real science and Professors Alan Lightman Anita Desai and David Thorburn for all the distractions in fiction literature and film Although I ve never met him Douglas Adams deserves recognition for impressing upon me the subtle wisdom of cynicism at a very young age which was further cultivated by Kurt Vonnegut and Mark Twain among others and special thanks is due to The Clash and Rancid for creating the quality punk rock albums that provided much of the acoustical accompaniment to this work On that note while
58. Swing components that would be necessary to make pop up menus which are only supported in version 1 2 implementations of the Java Runtime Environment Netscape and Microsoft have been lackadaisical in supporting this technology and consequently it is only now beginning to appear in the newest browsers 5I used an off the shelf open source implementation of the Javascript menus created by Gary Smith of the Netscape corporation The source code and documentation of Smith s Javascript menu 84 generates the Javascript that sets up and creates the menu objects It should be used to place this code somewhere between the header and the body of the document that is output by the content handler For example lt HTML gt lt HEAD gt lt TITLE gt The MIT City Scanning Project Dataset lt TITLE gt lt HEAD gt lt String strBGColor FFFFFF strHeadingColor 0066ff strColC olor 7ebffc gt lt jsp useBean id curPathBean scope request type NavPathBean gt lt int n j Branch pBranches BranchClass pBranchHistory curPathBean getBranchHistory String pHistory curPathBean getHistory AttributeMapBean curVtxAttrsBean curPathBean getCurVtxAttrs hr lt Create the Javascript menu objects gt lt j BranchHistoryScripter insertMenusScript new PrintWriter out curP athBean gt lt BODY BGCOLOR lt strBGColor gt gt As the example shows the BranchHistoryScripter gets all
59. Type maxOccurs unbounded gt lt xs complexType gt lt xs sequence gt lt attribute definitions need to come before gt lt signatures gt lt attributes block gt lt xs element name attributes minOccurs 0 gt lt xs complexType gt lt xs sequence gt lt xs element name int type intType minOccurs 0 maxOccurs unbounded gt lt xs element name String type stringType minOccurs 0 maxOccurs unbounded gt lt xs element name stateVar type stateVarType minOccurs 0 maxOccurs unbounded gt lt xs sequence gt lt xs complexType gt lt xs element gt lt signatures gt lt xs element name Signature type xs string maxOccurs unbounded gt lt xs sequence gt lt xs attribute name name type xs string use required gt lt xs complexType gt lt xs element gt lt vertex gt lt xs element name vertex maxOccurs unbounded gt lt xs complexType gt lt xs sequence gt lt xs element name parent min ccurs 0 gt lt xs complexType gt lt xs attribute name name type xs string use required gt lt xs complexType gt lt xs element gt lt xs sequence gt lt xs attribute name name type xs string use required gt lt xs attribute name type type xs string use required gt lt xs complexType gt lt xs element gt lt xs sequence gt lt xs complexType gt lt end data model type gt 160 lt attri
60. a non standard output stream to the OutputStreamServer which makes it possible to use the OutputStreamServer as some kind of remote logging utitily Any number of clients can connect to the OutputStreamServer which maintains a table of all connections to it and the OutputStreamServer is thread safe To use the OutputStreamServer for debugging JSPs and servlets I have cre ated another simple servlet called the MonitorServlet The MonitorServlet runs in the same webapp as the servlets and JSPs to debug and presents the user with the CGI form shown in Figure 7 11 This form provides some controls for starting and stopping an OutputStreamServer and when requested redirects the standard out and and standard error streams of the virtual machine in which it is running 138 Serviet and JSP Monitor Connect to monitor server Figure 7 11 The MonitorServlet to the OutputStreamServer All of the JSPs and servlets in the same webapp as the MonitorServlet will share the same virtual machine and therefore this redi rection of the output stream will affect all of them as well To test out one of the servlets or JSPs the developer just connects to the OutputStreamServer by click ing on the Connect to monitor server link which opens a telnet connection to the QutputStreamServer automatically on the specified port Then the developer types the URL into the form and presses go and the MonitorServlet forwards the developer to
61. a valid path relative to webapp to test break action if strTestURL charAt 0 pw printin lt P gt Path must begin with break action getServletContext getRequestDispatcher strTestURL forward request response 177 yelse if strAction equals Start Server start_server synchronized m_serverMutex if m_0SServer null need to initialize server get the port number from form data String strPort request getParameter port if strPort null pw printin lt P gt Can t start server you must provide a valid port number break start_server try m_nPort Integer parseInt strPort catch NumberFormatException e pw printin lt P gt Can t start server strPort is not a valid port number break start_server m_OSServer new OutputStreamServer m_nPort Servlet and JSP Monitor end if m_OSServer null if m_OSServer alive pw printin lt P gt Can t start server it s already running on port m_0SServer getPort break start_server if m_0SServer getPort m_nPort need to create a new one to change port m_OSServer new OutputStreamServer m_nPort Servlet and JSP Monitor m_OSServer start save a reference to old system out m_oldSysOutStream System out redirect system out to server System setQut new PrintStream m_OSServer pw printin lt P gt Serv
62. aLink Server API so these alternative content handlers would only be able to display content that doesn t make use of any of the data forwarded by the DataLinkServlet 4 2 The Format of DataLinkServlet Requests Before we can begin discussing the interaction between content handlers and the DataLinkServlet it is important to make note of a few of the essential pieces of a DataLinkServlet request Requests can be made to the DataLinkServlet using the GET or POST protocols A request in the GET format has the general form outlined below http city lcs mit edu 8080 data2 DataLink view node amp dataset all nodes fixed amp nodeld 75 amp history dataset all_nodes The first parameter in this request URL view specifies the name of the vertex to select The history parameter specifies the list of vertices that have been visited along the current navigation path delimited by the character not including the currently selected vertex which in this case is node The rest of the parameters are the names and values of attributes The client can specify additional parameters in a DataLink Server request using the same name value construct that is illustrated in this example The DataLink 1The name of the DataLinkServlet has been been shortened in this URL to DataLink using a customization parameter of the webapp which deploys the DataLink Server The reader unfamiliar with JSP and Java servlets might not know that the standard API for the
63. abstraction of the physical data The topmost abstraction barrier is created by the presentation layer which is the interface that is visible to the user and is responsible for formatting the data provided by the lower abstraction layers into HTML or other web browsable content The next level beneath the presentation layer is the navigation layer which is where the navigation hierarchies are defined The navigation layer describes what data should be visible to the presentation layer at any given point in the navigation hierarchy It also describes where a user can go to next from the current location in the hierarchy The navigation layer has a contract with the presentation layer such that it will only provide navigation options for data which is physically present In order to ensure that this covenant is fulfilled the navigation layer interfaces with the data layer in order to resolve the mappings between the virtual data abstractions in the navigation layer and real physical data in the dataset The data layer describes the physical hierarchy of the data and defines the attributes of each type of datum which might be present in the dataset Attributes are bits of information such as image resolution source sensor date of acquisition or any other properties of which should be presented as selections that the user can use to locate specific data Figure 2 1 depicts the interactions between the three abstraction layers of the DataLink Server As t
64. am strNavVtxName Nav vertex of requested resource param strDataVtxName The name of a data vertex for the type of resource requested param attrMap A set of name value pairings that define the attributes of the selected resource can lt b gt not lt b gt be null return a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt since JDK1 0 public static final InputStream openResourceStream String strNavVtx Name gt String strDataVtx Name SimpleNamedValue Map attrMap throws IOException Requests the resource at nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt from x the DataLink Server and returns it as an InputStream param strNavVtxName Nav vertex of requested resource 156 param attrMap A set of name value pairings that define the attributes of the selected resource can be null return a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt w
65. ame root type dataset gt ne G dataset gt dotsetName gt lt name opus number gt p lt name composer lt name COMPOSER Fa ee type composer_dir gt type composer_dir gt ene composerName _ lt name pieces um type mp3 gt lt name MUuSIC FILES Ly lt name BlO_DIR type music_di gt s type bio_di gt lt name portrait lt name bio f F type portait_img gt type bio_doc gt lt name PIECES lt name BIOGRAPHY lt name PORTAT type mp3 gt type bio_doc gt type portait_img gt opusNum o pleceName Figure 2 2 A simple set of models for a classical music library each composer in the library Each composer directory contains two subdirectories One is a biography directory which holds a document about the life of the com poser and an image of the composer s head The other is a music directory which contains audio files of pieces written by the composer The navigation model re maps this directory structure to a slightly different set of hierarchies The user can begin browsing the dataset by selecting the name of a composer causing the system to visit the composer vertex Once that selection is made the user will see a listing of pieces penned by the selected composer as well as a biographical sketch and a mug shot Clicking on the name of a piece will then download that recording An alternative route that the
66. ame of the nav vertex to view 149 param attrMap A set of name value pairings that define the attributes of the selected vertex return The URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which may be null x since JDK1 2 public static URL formatViewRequestURL String strNavVtxName NamedValueMap attrMap Returns the URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which must lt b gt not lt b gt be null This method is intended for use in applications where Java 1 2 may not be used param strNavVtxName Name of the nav vertex to view attributes of the selected vertex param attrMap A set of name value pairings that define the return The URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which must lt b gt not lt b gt be null since JDK1 0 public static URL formatViewRequestURL String strNavVtxName SimpleNamedValueMap attrMap Returns the URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attrib
67. an curVtx AttrsBean throws IOException Requests the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt from the DataLink Server and returns it as an InputStream param strNavVtxName Nav vertex of requested resource param strDataVtxName The name of a data vertex for the type of resource requested param attrMap A set of name value pairings that define the attributes of the selected resource can be null x x return a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt since JDK1 2 public static final InputStream openResourceStream String strNavVtx Name 159 String strDataVtx Name NamedValueMap att rMap throws IQException Requests the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt from the DataLink Server and returns it as an InputStream This method is intended for use in applications where Java 1 2 may not be used par
68. anywhere in the world to browse is dataset This web interface included a set of visualization tools that allow viewers to examine the quality of the data in various post processing stages and learn more about the process by which the data is acquired Bodnar 2001 This thesis outlines extensions and additions to the pre existing web interface in order to make it a powerful tool for analyzing the data These extensions make the web interface an easy to use exploration tool for members of the lo general public to browse the data and learn more about the project as well as a useful tool for members of the City Group to validate the data evaluate the performance of the various processing stages and diagnose problems with the acquisition process This section outlines briefly what the reader needs to know about the image ac quisition process and the pre existing web interface For more information about the data or the image acquisition and post processing stages please refer to Calibrated Registered Images of an Extended Urban Area Teller et al 2001 The previous ver sion of the web interface is documented in A Web Interface for a Large Calibrated Image Dataset Bodnar 2001 1 1 1 Image Acquisition and Post processing Stages Pose images are collected using a robotic sensor which consists primarily of a digital camera mounted on pan tilt head De Couto 1998 The sensor is equipped with an on board GPS receiver inertial motion sens
69. ature it gives precedence to data with the signature that appears first in the lt state gt declaration so for nodes with both all_trans and 5As new data is acquired the post processing pipeline is often reapplied to the entire dataset in order to take advantage of any new global features such as vanishing points and baselines that have been discovered in the new nodes 92 Abstract State Literal State unbound null null extracted from auto_rot datum or rotated auto_rot requesting auto_rot data in request for rotated data rotated null Table 3 2 Valid state pairings listed by example trans_set4 directories the DataLink Server will return the contents of the all trans directory when it receives a request for camera files in the translated abstract state of pose refinement It is possible to make requests to the DataLink Server which override this behavior in order to obtain camera files from only the trans_set4 directories or vice versa State Attributes Have Abstract and Literal Values State attributes can hold two types of values simultaneously an abstract state and a literal state The literal state of a state attribute is the particular state signature that it is bound to When state attributes are extracted from physical data they will have both a literal state which was obtained from the data itself and a corresponding abstract state When making requests to the DataLink Server the client may only be interested in the
70. avVtxName String strDataVtx Name NamedValueMap attr Map x rt amp amp F KF KF HF KF F Returns a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt This method is intended for use in applications where Java 1 2 may not be used param strNavVtxName Nav vertex of requested resource param strDataVtxName The name of a data vertex for the type of resource requested 152 param attrMap A set of name value pairings that define the attributes of the selected resource can lt b gt not lt b gt be null 2 return a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt since JDK1 0 public static String formatViewResourceRequest String strNavVtxName String strDataVtxName attrMap IIIIT III TIII III TI III III formatViewResourceRequestURL LL Returns a URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code
71. ayer to Vz in the navigation layer When the user selects the new vertex the navigation layer must tell the presentation layer what navigation options are available and what data is visible to the user from the new vertex It does this by examining each of V s children as illustrated by the green arrows These child vertices may map to some components of the physical hierarchy What data they map to is dependent on two things 1 the data that actually exists in the dataset and 2 the attributes of the current navigation path 31 Recall that the data layer defines attributes for each datum in the dataset At tributes are variables whose values get assigned at runtime either by selections made by the user or by a pattern matching process that occurs in the data layer which identify data in the dataset For example nodes in the City Scanning Project Dataset are identified by an ID number If V2 in Figure 2 1 was used to represent nodes the content hander for V would probably display links of the form Node 1 Node 2 and so on for every node that exists in the dataset provided the set of nodes visible to the user at this point was not constrained by other attribute values accumulated along the user s current path through the navigation graph The user would move to V in the navigation hierarchy by selecting one of these links When the user makes one of these selections a value for the node ID attribute would be assigned For e
72. bute types gt lt int attribute gt lt xs complexType name intType gt lt xs attribute name name type xs string use required gt lt xs complexType gt lt string attribute gt lt xs complexType name stringType gt lt xs attribute name name type xs string use required gt lt xs complexType gt lt state attribute gt lt xs complexType name stateVarType gt lt xs sequence gt lt xs element name State max ccurs unbounded gt lt xs complexType gt lt xs sequence gt lt xs element name signature type xs string maxOccurs unbounded gt lt xs sequence gt lt xs attribute name name type xs string use required gt lt xs complexType gt lt xs element gt lt xs sequence gt lt xs attribute name name type xs string use required gt lt xs complexType gt lt nav model type definition gt lt xs complexType name navModelType gt lt xs sequence gt lt xs element name vertex maxOccurs unbounded gt lt xs complexType gt lt xs sequence gt lt children gt lt xs element name child minOccurs 0 maxOccurs unbounded gt lt xs complexType gt lt xs attribute name name type xs string use required gt lt xs complexType gt lt xs element gt lt properties gt lt xs element name property minOccurs 0 max ccurs unbounded gt 161 lt xs complexType gt lt xs attribute name name type
73. butes of the currently selected vtx can be obtained from the NavPathBean forwarded by the DataLinkSerlvet param pHistory An array containing the names of the n vertices visited thus far which pHistory 0 pointing to the first vertex the root and pHistory n pointing to the latest return A String containing the URL for a view request for the Supplied arguments since JDKi 2 public static String formatViewRequest Branch branch string strNavVtxName AttributeMapBean curVtxAttrs Bean x String pHistory Returns a String containing the URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which may be null param strNavVtxName Name of the nav vertex to view 146 param attrMap A set of name value pairings that define the attributes of the selected vertex request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by return a String containing the URL for a propererly formated x lt code gt attrMap lt code gt which may be null since JDKi 2 public static String formatViewRequest String strNavVtxName NamedValueMap attrMap xx Returns a String containing the URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by t
74. ch like a directory listing or it may incorporate Java applets Javascript DHTML or other means to make the data interactive As Figure 2 3 demonstrates the content handlers sit on top of the DataLinkServlet architecture Figure 2 4 shows the document the content handler for the composer vertex in the classical music dataset might produce for the branch dataset classical composerName Brahms As the figure shows the data for the portrait and bio vertices are incorporated into the output document itself whereas hyperlinks whose action is to dispatch a query to the DataLinkServlet for a particular branch of the pieces vertex are provided for the user to download the audio files Figure 2 5 shows the output of the content handler for the node vertex which maps to a node directory in the City dataset This content handler makes use of many interactive elements such as Java applets for viewing the spherical texture a clickable map applet for browsing through the other node branches and cascading Javascript menus which allow the user to jump to sibling vertices of those in the current navigation path 6The biographical text used in this example is an excerpt from Music An Appreciation by Roger Karmien pp 256 258 Karmien 1999 40 Johannes Brahms Works Sonata for Clarinet and Piano no 1 op 120 Sonata for Clarinet and Piano no 2 op 120 Biography 5 Johannes Brahms 1833 1897 was a romantic who breathed new life i
75. coordinates 6 2 1 Design Principles of GCTL The examples in the previous section illustrate several of the software design principles that went into the GCTL API Briefly these are e Though it may require learning more classes and methods it is better to create simple classes which are designed to do one specific task well than to try to pack too much power into a generic portal 105 e Associate each transform with a frame of reference e Use a naming convention the specifies to and from coordinate types explicitly e Coordinate types should have their own data types Some of the advantages of embracing these conventions are e The system is easy to learn because of the teleological naming convention e The library is extensible to add new transforms all one has to do is extend one of the simple base classes or create a new one There is no need to modify or understand the implementation of the existing API e All of the tools associated with a particular coordinate system or frame of reference are localized to one computation structure tailored for that task This in turn facilitates chaining multiple transformations 6 3 GCTL Understands Five Types of Coordinates Before describing how to perform geospatial coordinate transformations with GCTL it is worth reviewing the types of of geospatial coordinate systems GCTL can work with GCTL knows about six different types of coordinate systems whose definitions f
76. cture is essential to achieve the high degree of customizability provided by the DataLink Server All of these components interface with one central appli cation the DataLinkServlet which handles the processing of requests Only the DataLinkServlet itself and the software behind its API are off limits to developers All of the other components are created or extended by web developers to implement a custom web application The idea is that the web developers only have to describe how the web site should be laid out as well as what functionality it should provide and the DataLink Server infrastructure should take care of the rest The follow 33 ing sub sections describe each of the essential components of the DataLink Server in detail 2 2 1 Data and Navigation Models Define the Abstraction The abstraction function provided by the DataLink Server is defined by a pair of directed graphs called the navigation model and the data model Section 2 1 and Figure 2 1 have already given some sense of the nature of the navigation model As we have seen so far the navigation model consists of vertices which represent locations in the virtual navigation hierarchies Most of these vertices really represent some type of datum in the physical dataset This data type may be a resource such as an image or a data file type or it may be a type of vertex in the physical data hierarchy such as one of the node directories that contains all of the data for a
77. d have the following set of generalized attribute bindings refinement rotated resolution half The algorithm that performs branch merging and generalization is O n where n is the number of branches before merging so there is a small cost associated with merging when the number of branches is large When it is known that duplicate branches can never occur for a particular navigation vertex the merge_duplicates parameter can be set to false to disable the merging process for that vertex and im prove the performance of the DataLink Server The node vertex of the City dataset s navigation model is a good example of this because every node directory is identified by a unique ID number and therefore there can never be duplicate branches for the node vertex enumerate_history In addition to the branches for each child of the selected vertex the DataLink Server also returns sibling braches for all of the vertices in the history of the current navigation path These can be formatted by using the BranchHistoryScripter tool in the DataLink Server API to create context menus which allow the user to jump back to earlier points in the navigation path or to quickly make an alternate selection for a previously visited vertex Enumeration for individual vertices can be turned off by setting the enumerate_history parameter to false This can be used to reduce computation or to prevent the display of menus for vertices which have may have a v
78. der will learn about the GCTL API This will include an overview of the basic data types and an exposition of the all of the different types of transformations which can be carried out using GCTL This section will illuminate as much as the reader should need to know about geospatial coordinate systems to understand the basics of GCTL its implementation and it s usage 6 4 1 GCTL Data Types There are two data main structures for representing coordinates in use in GCTL Vector3D and LLA Vector3D is used for any 3D Cartesian coordinate system which includes geocentric and local grid coordinates Cartographic projections each have their own special data types For example US State Plane coordinates are repre sented by a USSP data type The data types for cartographic projections are re ally just typedefed Vector3D classes with names that evoke the coordinate system they represent For cartographic coordinates the third component of the underlying Vector3D type is implicitly 0 An LLA is used to store coordinates as latitude longitude and altitude Coor dinates of this form are called geodedic coordinates They may also sometimes be referred to as a latitude longitude height LLH coordinates The precise definition of altitude varies with the coordinate systems and surveying methods in use but for all of the coordinate systems used in GCTL altitude refers to the perpendicular distance of a point from the surface of a reference ellipsoid
79. dinates of the current iteration If the user has selected a check box labeled scale located beneath the viewing windows hash marks perpendicular to the epipolar line labeled with the distance from the node s eye are drawn for each of the z y sample points see Figure 7 6 unless a maximum number of hash marks 30 in this implementation have already been plotted This prevents the epipolar line from becoming too crowded with tick marks near the vanishing point when AX and AY become very small At this point 127 Node 335 Node 337 iA u 4 a Figure 7 6 An epipolar viewing of Building 18 with distance markings on the epipoles the step size also begins to increase by multiples of 10 instead of 2 The algorithm terminates when the step size passes a threshold value or a maximum number of iterations have been completed This algorithm is good because it is reasonably cheap computationally speaking and the simple heuristics for increasing the step size give the distance markings on the epipoles a nice intuitive logarithmic scaling with integer distance values Another alternative way to achieve uniform spacing of the distance markings in screen space would be to interpolate evenly in screen spacing while mapping the screen coordinates to epipolar coordinates using a technique borrowed from texture mapping sS t ti ot rr he ti 7 a Olas a Here s is a point on the parametrized projected line s
80. ds up display as shown in Figure 7 4 In this image the user has activated the heads up display by clicking on a check box labeled compass directly beneath the mosaic viewer not depicted As the figure shows the heads up display shows a gold target to denote the current viewpoint as well as compass markings at the top of the window which indicate the horizontal viewing direction in degrees and hash marks 125 tif teri E A A be it Bias a i Hh aig i Zu en aL 35 Au 45 mi N uch a Ti Sonal gt iT Figure 7 4 The heads up display along the left hand side which indicate the degrees of elevation The 0 90 180 and 270 degree hash marks for the compass heading are labeled with N E S and W respectively By default the node viewers always start with the viewpoint aimed at the direction of due North This in conjuntion with the map viewer which is oriented such that North is the vertical direction on the page is meant to give the user some sense of direction while browsing the dataset Edge and Point Features Can Be Superimposed on the Node s View The node viewers have also been augmented with edge and point i e edge intersec tion visualizations which can interactively display the features detected during the post processing stages The edge feature visualization can be activated by clicking on a check box beneath the node viewer which is labeled edges The
81. e part of this process which is worth discussing before delving into the details of how the DataModel handles its part of the transaction however is the branch merging and generalization operation that is performed in Step 4 8 Hashing Makes Merging More Efficient If the NavModel did a straightforward item by item comparison of each of the branches of c to check for duplicates the time requirement for this process would be O n where n is the number of branches This could be a serious bottleneck because n can often be very large This can be reduced to an O n problem however by applying a hash function to each of the branches in order to created a hashed set Whenever two branches are the same they will always generate the same hash code and if the hash function is a good one then most of the time when branches are different they will generate different ones Thus the NavModel can merge the branches quickly by placing each one in the hashed set and watching for hash code collisions Whenever there is one the NavModel compares the two branches and if they are indeed the same merges and generalizes them Branches are compared by name and the hash function that is used to place them in the hashed set is the standard Java hash function for strings 2The hash code for null strings is defined to be 0 94 A string 31275 T 312 ER 5 1 In order to reuse this hash function for merging branches based on the value of a d
82. e and longitude geo centric coordinates which are used by satellites and GPS systems cartographic co ordinates which are used in maps and local grid coordinates which are often used for surveying Prior to the existence of this library there were no static libraries that programmers could use to perform all of these tasks together This is a major grievance because transformations between three or more different geospatial repre sentations may often need to be done in tandem One extreme example occurred in the City Scanning Project when over the past year the team developed an inter 101 est in exchanging information between the City dataset s geospatial images of MIT s campus and a campus base map maintained by the Department of Facilities This requires transformations which span all four of the geospatial coordinate domains and is now accomplished by GCTL 6 1 Problems With Existing Packages In order to appreciate some of the advantages of the software design principles that went into the interface of GCTL it is worthwhile to take a look at some of the prob lems with existing packages that motivated them Aside from the lack of a single unified API for geospatial coordinate transformations there are several other deficien cies associated with the other pre existing coordinate transformation packages 1 There is a dearth of programmatically accessible coordinate trans formation libraries Most of the pre existing
83. e for mat of the history parameter or how it is maintained since this is the job of the DataLinkDispatcher The developer of content handlers just needs to be aware of the history s purpose in order to understand when to use DataLinkDispatcher routines that invoke it and when it is safe to use the ones that do not 4 3 Data Structures for Content Handlers The chief purpose of content handles is to format the data served by the DataLinkServlet in a useful and pleasing way Therefore to write content handlers one needs to be familiar with the data structures that are used by the DataLink Server to encapsulate the data it communicates to them This section outlines all of these important data structures briefly 4 3 1 The DataLinkServlet Communicates to Content Han dlers Using the curPathBean All of the information about branches and navigation vertices that the DataLinkServ let provides for content handlers is communicated using a Java Bean called the curPathBean The DataLinkServlet passes this bean to the content handler of a navigation vertex when that vertex is selected This curPathBean is an instance of the NavPathBean class This class has a number of routines for use in content handlers which can be used to obtain information about the current navigation path 2Any reference book on JSP and Java Servlet technology such as Proffesional JSP Brown et al 2000 should be able to provide a more thorough treatment about the l
84. e logical hier archies of the data any of which may be navigated by the user to arrive at the same data e Content Managers make it possible to create a modular customized facade for the front end of the interface e The DataLink Server handles updates in real time In addition to these advantages the DataLink Server also improves the efficiency of the web interface by employing a caching strategy for large blocks of commonly accessed data such as the locations of the nodes to be plotted by the Map Viewer and performs user auditing to help the City group keep tabs on the site s usage The complete details about the caching strategy and user auditing as well as the inner workings of the DataLink Server itself and the XML based modeling language are discussed in Chapters 2 4 1 2 2 The Geospatial Coordinate Transformation Library The City Scanning Project dataset uses a local tangent plane LTP coordinate system for all of its position and orientation data This coordinate system is defined by a right handed set of axes with the x axis pointing east the y axis pointing north and the z axis pointing up normal to the Earth with units in meters The origin is located at the position of the GPS base station antenna on the roof of building NE43 in Tech Square that the City group uses to obtain differential GPS measurements of the nodes placement This coordinate system is used to provide the highest degree of precision to po
85. e navigation model block of the model file which describes the interconnections between a network of navigation vertices The basic outline of a navigation model looks like this lt navModel gt lt navigation vertex declarations go here gt lt root name name of root navigation vertex gt 08 Navigation Model lt name classical lib tyoe dataset gt lt name opus number gt lt lt name composer type composer_dir gt lt name pleces lt name portrait lt name bio type mp3 gt type portait_img gt type blo_doc gt Figure 3 2 Navigation model of the classical music library lt navModel gt 3 4 1 Defining Navigation Vertices Enumerating a list of all the features involved in creating a navigation model would be tedious to read and far less trenchant than a series of demonstrative examples so in that light we will continue this discussion of navigation models by building one for the classical music library The basic structure of the navigation model we will build is shown in Figure 3 2 After detailing how this navigation model is put together this section will show how additional parameters of navigation vertices can be used to fine tune the DataLink Server s exhibition of this model The best way to design a navigation model is from the bottom up starting with vertices for the files to make available lt navModel gt lt vertex name pieces ty
86. each vertex in the chain for the requested data vertex Remember that I mentioned that data vertices are currently only allowed to have one parent 98 2 1 Evaluate the signature for the current vertex and append it to the base path 3 Serve the data at the accumulated path in Chapter 3 and hence here they can have only one chain Adding support for multiple parents would require some modification to the view resource request because the attribute bindings by themselves might be ambiguous 99 100 Chapter 6 The Geospatial Coordinate Transformation Library The second major contribution of this thesis in addition to the DataLink Server is the Geospatial Coordinate Transformation Libraray This coordinate transformation library was conceived out of a desire to unify data in the City Scanning Dataset which is represented in the City group s internal coordinate system with the vast amount of geospatial data such as maps and satellite images that exists in the world but it grew into a much broader vision to encapsulate all of the most commonly used geospatial coordinate transformations into one powerful customizable and compre hensible library which can be reused in any software application The Geospatial Coordinate Transformation Library GCTL supports transforma tions between the four most commonly used representations of points on the Earth s surface geographic coordinates which are expressed in latitud
87. eat Sail 1966 Massachusetts Institute of Technology Cambridge 20 Mosaic Viewer planar projection of spherical mosaic Show compass _ baselines edges I intersections vanishing points To use the Mosaic Viewer click on the image above and drag if to the desired viewpoint You can also rotafe fhe viepoint by using fhe arrow keys or by dragging and dropping the POY marker the orange target The Zand w keys zoom in and out respectively Figure 1 2 A screen shot of the Mosaic Viewer enhanced with some of the extensions described in this thesis 21 GG Node 324 Node 326 Legend location of other node POV drag drop to move point of projection of epipolar line 177 epipolar line with scale in m edge feature edge intersection vanishing point baseline direction saturation threshold pose estimate Use translated pose estimate scale I compass 7 baselines ShowFeatures 7 edges IT intersections vanishing points Status Figure 1 3 A screen shot of the Epipolar Viewer comparing the geometry of two views of The Great Sail 1 2 Overview of the Extended Interface The extended interface has roughly the same look and feel as the original pre existing interface It allows users to browse the dataset in nearly the same fashion as before and view the data using improved versions of the visualization applets described in the previous s
88. eb portal when these data are accessed A simlar caching strategy could also be applied to the 137 vanishing points data used on the same page 7 3 User Auditing and Debugging In order to keep tabs on the usage of the web interface as well as to help identify problems with it I have set up a user auditing system which logs the IP address of each client that accesses the web interface combined with information about what resource servlet JSP or HTML document they requested This information is stored to the Tomcat servlet container s log file for the City data webapp I have also created another pair of remote auditing and debugging tools the MoniterServiet and the OutputStreamServer These tools attack one of the major difficulties associated with developing JSPs and Java servlets These applications run on a remote server which makes them difficult to debug Many times a developer would simply like to output some printf or in this case System out println statements to the console in order to get an idea of the buggy program s behavior These statements will simply be redirected to some console on the remote server however and the developer will never be able to see them The OutputStreamServer solves this problem by allowing a developer to redirect one or more of the standard output streams to a simple server that forwards anything output to the stream to a remote client via a TCP IP connection It is also possible to redirect
89. ection In addition to enhanced functionality the new web interface is backed by a new powerful and versatile navigation infrastructure whose core is a Java servlet package called the DataLink Server A C static library for transforming Earth coordinates has also been added to the visualization toolkit allowing the dataset to be used in conjunction with data from outside sources expressed in almost any geographic geocentric or cartographic coordinate system This section gives a brief overview of the DataLink Server the coordinate transformation library and other minor extensions and improvements that have been made to the pre existing interface all of which will be described in greater detail in subsequent chapters 1 2 1 The DataLink Server For the pre existing web interface the abstraction between the organization of the data presented to the user and its arrangement on the physical disk is provided by a few Perl scripts These scripts traverse the directory tree of the physical data and generate symbolic links to the relevant files They must be run prior to deployment of the web site The symbolic link trees re map the complicated directory structure of the physical data to a simpler hierarchy that is more suitable for browsing on the web Bodnar 2001 Although re mapping the data in this way effectively provides the desired abstraction the symbolic links are problematic in that they must be regenerated every time there are additions
90. edProperty name opusNum source mp3 gt lt param name supress_incomplete_leaves value true gt lt vertex gt When the suppress_incomplete_leaves parameter is set the DataLink Server will supress the display of branches for leaf navigation vertices whose inherited at tribute values have not been completely specified by the current navigation path Therefore with the suppress_incomplete_leaves parameter set when the DataLink Server is at the opus number vertex it will not display branches for audio files unless it knows values for both the datasetName and composerName attributes merge_duplicates Sometimes the order of two attributes can be switched by some means other than a derived property One example occurs in the City dataset for the attributes for image resolution and state of post processing refinement These 66 attributes derive their values from the names of directories in the City dataset In the physical hierarchy of the City dataset the resolution directories are the parents of the post processing directories The navigation model for the City dataset allows the selection of these attributes to be reversed by cross referencing the navigation vertices associated with them lt navModel gt lt vertex name pose refinement type pose_refinement gt lt child name resolution gt lt vertex gt lt vertex name resolution type resolution gt lt child name pose refinement gt lt vertex gt
91. employ one of many named ellip soids GUTL also has internal representations of these which are listed in Table 6 2 108 Ellipsoid GCTL Class generic ellipsoid Ellipsoid Airy 1830 Airy1830_Ellipsoid Australian National Australian_National_Ellipsoid Bessel for Ehiopia Indonesia Japan and Korea Bessel1841_BR_Ellipsoid Bessel for Namibia Besse11841_BN_Ellipsoid Clarke 1866 Clarke1866_Ellipsoid Clarke 1880 Clarke1880_Ellipsoid EE W Malaysia and Singapore 1948 EE_Ellipsoid ED W malaysia 1969 ED_Ellipsoid Everest Everest_Ellipsoid GRS80 GRS80_Ellipsoid Helmert 1906 Helmert906_Ellipsoid Hough 1960 Hough1960_Ellipsoid India 1830 India1830_Ellipsoid India 1956 India1956_Ellipsoid International 1924 International1924_Ellipsoid Krassovsky 1940 Krassovsky1940_Ellipsoid Modified Fisher 1960 Modified_Fischer1960_Ellipsoid South American 1969 South_American1969_Ellipsoid WG972 WGS72_Ellipsoid WGS84 WGS84_Ellipsoid Table 6 2 GCTL Ellipsoid Classes New datums classes for use with GCTL can be created by extending the generic GeoDatum base class as follows class New_Datum public GeoDatum public New_Datum GeoDatum Some_Ellipsoid deltaX deltaY deltaZ rotx rotY rotz scale 3 Here the additional arguments after the ellipsoid are the elements of a 4 x 4 homogenous matrix which defines the change of basis from the new datum to WGS84 6WGS84 is the common datum that GCTL uses to link all datu
92. ensing org 2000 remotesensing org 2000 GeoTIFF 7 Apr 2002 lt http www remotesensing org geotiff geotiff html gt Ritter and Ruth 2000 Ritter N and Ruth M 2000 GeoTIFF Format Spec ification GeoTIFF Revision 1 0 Technical report 7 Apr 2002 lt http www remotesensing org geotiff spec geotiffhome html gt Schildt 1995 Schildt H 1995 C The Complete Reference Osborne McGraw Hill Berkely California second edition Smith 1998 Smith G 1998 The Javascript Menu Object View Source Magazine 16 May 2002 lt http developer netscape com viewsource gt Sperberg McQueen and Thomson 2002 Sperberg McQueen C M and Thomson H 2002 XML Schema Technical report The World Wide Web Constortium 11 May 2002 lt http www w3 org XML Schema dev gt Teller et al 2001 Teller S Antone M Bodnar Z M Bosse M Coorg S Jethwa M and Master N 2001 Calibrated Registered Images of an Extended Urban Area In Proc IEEE CVPR lt http graphics lcs mit edu seth pubs gt 183 The Apache Software Foundation 2001 The Apache Software Foundation 2001 The Apache XML Project 11 May 2002 lt http xml apache org gt The World Wide Web Consortium 2002a The World Wide Web Consortium 2002a Extensible Markup Language XML 11 May 2002 lt http www w3 org XML Schema dev gt The World Wide Web Consortium 2002b The World Wide Web Consortium 2002b Wor
93. er The action of the navigation layer is also 1The actual format of the the links is up to the content handler 32 minimal in this case It simply requests the appropriate data from the data layer and then passes it along to the user 2 1 1 The DataLink Server Can be Accessed Programatically For simplicity the previous discussion referred to the client of the DataLink Server as the user However requests to the DataLink Server are made using the GET or POST protocols which means that the DataLink Server can also be accessed programatically like a CGI application This gives content handlers or other web applications the ability to locate data within a dataset behind the scenes in order to generate more dynamic content without any code for locating physical resources being hard wired into the actual web content The DataLink Server package includes a Java API which can be used to create properly formatted requests for items in the navigation hierarchies given an appropriate set of attributes Section 4 4 and Appendix A give the complete details of the API for formatting requests to the DataLink Server As we ll see in Section 5 2 the typical format for DataLink Server requests using the GET protocol looks like DataLink view node amp dataset all nodes amp nodeld 11l amp history dataset all_nodes 2 2 Components of the DataLink Server The DataLink server is comprised primarily of five discrete components This mod ular archite
94. er type composer_dir gt lt child name opus number gt lt child name pieces gt lt child name portrait gt 10 lt child name bio gt lt branchName gt s s surname firstName lt branchName gt lt branchClassName gt Composers lt branchClassName gt lt contentHandler gt composer jsp lt contentHandler gt lt vertex gt The contentHandler commands specifies a URL to the content hander for the vertex This URL must begin with a forward slash which specifies a URL relative to the home directory of the webapp where the DataLink Server is deployed 3 5 Chapter Summary That s pretty much all there is to the model file format Now that you know how to use the XML based modeling language to describe the organization of a physical dataset and the attributes of the data which should be navigable by writing a data model definition as well as how to describe the navigable hierarchies the behavior of the navigation layer by defining a navigation model you have mastered almost all you need to know to begin deploying powerful web interfaces to hierarchical datasets using the DataLink Server The next skill to acquire is the ability to write content handlers to provide a clean modular custum fa ade to the web interface provided by the DataLink Server This will be the subject of the next chapter 71 12 Chapter 4 Content Handlers This chapter describes the API tools the DataLink Server package provides
95. er 3 This chapter in conjunction with the previous chapter on modeling data and navigation using XML should provide the foundation necessary for a de veloper to begin using the DataLink Server The next chapter will focus on how the DataLink Server system actually works For a more in depth summary of the DataLinkDispatcher API be sure to check out Appendix A 86 Chapter 5 Operation of the DataLink Server This Chapter will describe how the core of the DataLink Server system works It will describe the algorithms and data structures used by the DataLinkServlet to service requests to the DataLink Server This will include a discussion of the DataModel and the NavModel the software implementations of the data layer and the navigation layer and will conclude with a walkthrough of a view request and a view resource request 5 1 Data Structures for Curbing Computational Complexity As has been discussed in Chapter 2 the software components that implement the data layer and the navigation layer are the DataModel and the NavModel By now the reader should be aware that the basic structure of these classes is a directed graph This section will expound the workings of these graph structures a little further One of the major items of concern in designing the DataLink Server was con trolling the computational complexity of the algorithms it employs for exploring the navigation graphs and locating data Finding specific vertices in each of
96. er scale than that of the DataLink Server and the coordinate transformation library Each of these is documented in Chapter 7 Among the extensions and refinements added to the pre existing web interface are e A compass visualization in the mosaic viewer and epipolar visualization tools e The Sphere View Port a small window that shows how points in the planar projection of the sphere texture in the mosaic viewer and epipolar visualization tools map to the cylindrical projection e Edge and point feature display in the mosaic viewer and epipolar visualization e Improvements to the epipolar visualization including distance markings on the epipoles e A Baseline visualization in the mosaic viewer epipolar viewer and the map viewer e A Vanishing point visualization in the mosaic viewer epipolar viewer and the map viewer e The use of a vector format map in the map viewer registered with the node data using the coordinate transformation library and support for arbitrary zoom factors lt http www remotesensing org geotiff geotiff html gt 27 28 Chapter 2 The DataLink Server The DataLink Server is designed to facilitate the development of an easily configurable interface for navigating a vast data store on the web It was created as a step toward fulfilling the desire for a system that abstracts the true organization of the data by providing a pliable set of virtual hierarchies which can be tailored to web navigation
97. er started on port m_nPort u fe end start_server synchronized m_serverMutex 178 else if strAction equals Stop Server stop_server synchronized m_serverMutex if m_OSServer null pw printin lt P gt No server running break stop_server System out println MonitorServlet User requested server shutdown m_OSServer stop reset system out System setOut m_oldSysOutStream pw printin lt P gt Sent server shutdown signal end if strAction null display the form data pw printin lt P gt lt A HREF telnet request getServerName m_nPort target _blank gt Connect to monitor server lt A gt pw println lt FORM METHOD post gt pw printin lt TABLE gt pw println lt TR gt pw printin lt TD gt Test URL lt TD gt pw printin lt TD gt lt INPUT TYPE text NAME testURL SIZE 60 gt lt TD gt pw printlin lt TD gt lt INPUT TYPE submit NAME action VALUE Go gt lt TD gt pw println lt TR gt pw printin lt TABLE gt pw printin lt TABLE gt pw printin lt TR gt pw println lt TD gt Port lt TD gt pw printin lt TD gt lt INPUT NAME port SIZE 6 VALUE m_nPort gt lt TD gt pw printin lt TD gt lt INPUT TYPE submit NAME action VALUE Start Server gt lt TD gt pw println
98. ere are of course limitless other ways that a coordinate transformation library capable of unifying data expressed City group s internal LTP coordinate system and the major coordinate systems used in mapping and surveying could be beneficial For this reason I have created the Geospatial Coordinate Transformation Library This coordinate transformation library is a static C library which can be linked to any of the City group s existing software or new applications which provides the capability to transform coordinates expressed in an arbitrary LTP coordinate sys tem such as the one used by the City group to any of the most widely used geo graphic geocentric and cartographic coordinate systems The coordinate transforma tion library could be used in conjunction with an open source library libgeotiff remotesensing org 2000 for exporting the images in the City Scanning Dataset to GeoTIFF Ritter and Ruth 2000 the image format which is becoming the de facto standard for encoding geospatial imagery Chapter 6 describes the Geospatial Coor An extensive source of information about the GeoTIFF format is the official GeoTIFF website 26 dinate Transformation Library in detail and supplies the reader with the background information necessary to use it effectively 1 2 3 Other Minor Extensions and Improvements This thesis also describes a number of other extensions and improvements to the pre existing web interface of a small
99. ere the data layer matches partial sets of attribute bindings to physical data specified by a data vertex chain This is done using a very simple recursive procedure applied to each vertex chain In addition to a vertex chain and the corresponding attribute set A the matching procedure also takes as arguments a base path and a start index 2 into the vertex chain Before making the entry level call to this procedure the data layer initializes the base path to be the root path to the dataset and sets 1 indez vp Here is an outline of this procedure 1 Loop over each vertex in the chain starting at v and terminate at vp 1 1 Try to evaluate the signature for v using the bindings in A If successful 1 2 1 3 append the evaluated signature to the base path decrement 2 and go back to Step 1 to process the next vertex Otherwise proceed to the next step signature evaluation will fail if required attributes are null Obtain a listing of all of the files and directories at the location specified by the base path For each datum in the listing attempt to bind the null valued bindings of v to the attributes of the datum If successful create a new attribute set A by inserting the new bindings into A and call this procedure again with 7 i 1 A and a new base path constructed from the evaluated signature of v appended to the old base path and add the results to a vector of bindings Vpindings
100. erived property the value of the property is compared in string form This eliminates the need to use a different hashing strategy for each of the different attribute types 5 3 2 How the Data Layer Fetches the Branches for a Data Type In Step 4 7 of the navigation layer s procedure for handling a select vertex request it requests all of the branches for the data type T of one of the selected vertex s children given a set of attribute bindings B from the data layer Here are the steps that the data layer takes to procure these branches 1 The DataModel looks in its table of data vertex definitions for all of the instances of type T and finds their locations in the graph by hashing 2 For each of the instances vp of type T that the data layer finds It follows the edges in the data model graph to build an ancestor chain from vp to a root vertex vp in the data graph 3 For each chain the data layer then adds any unresolved attributes of 2 to the set of attribute bindings Be to create a new set of attributes A some of which may be null valued and some which may not which will be used to match the request to physical data 4 Finally the data layer accumulates all of the possible bindings for each of the A and returns them all bundled as branches in a BranchClass to the navigation layer 95 How the Data Layer Matches Attributes to Physical Data Step 4 is the part of this procedure wh
101. erver does when it receives any request is to parse the request and determine what type it is To determine the request type the DataLink Server looks for the presence of the resource parameter If this parameter is not included in the request the DataLink Server treats the request as view request and proceeds by extracting the selected navigation vertex vy from the view param eter It then bundles up the set of attribute bindings Bin supplied by the request into a NamedValueMap for programmatic access by the NavModel the attributes are later translated into DataLink Server s native internal representation when more in formation about what data types they are to be associated with is known Finally the DataLink Server parses the history if one was supplied with the request into a convenient form and makes a select verter request to the navigation layer with all of 91 Presentation Layer Content Handlers JSPs Serviets HTML Applets hy 7 N serve data resolve a na u u 4 j _ J Data Layer Figure 5 1 Data paths through the DataLink Server this information 5 3 1 How the Navigation Layer Handles a Select Vertex Re quest Recall that the software implementation of the navigation layer is the NavModel class When this class receives a select vertex request it must examine navigation graph to see whether the selected vertex vy is an interior vertex or a leaf vertex This determi
102. ery large number of branches 68 Formatting Rules for Branches and Branch Classes The DataLink Server provides a couple of formatting options for branch names and branch class names A branch class is the set of all branches belonging to a particular navigation vertex In the default content handler view the branch class name is used for the heading placed above the branch listing for a particular navigation vertex As you can see from Figures 3 3 and 3 4 the default name for each branch class is simply the name of the corresponding navigation vertex A better branch class name can be provided for each navigation vertex by adding a branchClassName declaration to the body of the navigation vertex definition There is a similar command branchName which performs the same function for the names of branches Any valid string format expression of the same form as a data vertex type signature expression can be used for a branch name or a branch class name The following code example shows how to improve the formatting of the composer vertex lt modelDefs xmlns xsi http www w3 org 2001 XMLSchema instance xsi no NamespaceSchemaLocation http city 1cs mit edu 8080 data2 models mode ls xsd gt lt dataModel gt lt composer directory gt lt vertexType name composer_dir gt lt attributes gt lt string name firstName gt lt string name middleName gt lt string name surname gt lt attributes gt lt signature
103. ested reader can examine the exhaustive listing of the methods in the DataLinkDispater API that is located in Appendix A 4 4 1 View Requests vs View Resource Requests In and of itself the formatViewRequest method provides all of the functionality needed to make requests to the DataLink Server The other methods exist primarily for convenience and simply incorporate a few other tasks that are commonly associ ated with making requests to the DataLink Server such as opening an input stream The formatViewResourceRequest and formatViewResourceRequestURL methods are somewhat special however When the DataLink Server returns the resource branches corresponding to a leaf vertex in the navigation graph it has already done all of the work it will ever need to do to retrieve the resources associated with these branches it already had to find them once to determine whether they exist and what their attributes are The attributes associated with each branch are enough to completely specify its location so to find the resource associated with one of these branches all the DataLink Server has to do is ask the data layer to resolve the physical location for the resource directly from the attribute bindings no interaction with the navigation layer is necessary When the request is made with the standard view request format however the DataLink Server has no way of knowing that it is retrieving a resource without invoking a query to the navigation layer
104. f the node vertex which are consistent with the set of attribute bindings dataset all nodes fixed nodeId 75 The client can specify additional parameters in a DataLink Server request using the same name value construct that is illustrated in this example The DataLink Server treats any additional arguments to the request which do not correspond to attribute names as additional parameters for the content handler of the selected vertex and passes them on to the content handler without interference 5 2 2 View Resource Requests View resource requests ask the DataLink Server to serve a physical resource without invoking a query to the navigation layer In order to do this the attributes supplied in the request must completely specify the resource to retrieve History parameters and additional parameters have no effect on view resource requests and are simply ignored by the DataLink Server The following is an example of a view resource request http city lcs mit edu 8080 data2 DataLink view sphere amp resource SPHERE amp nodeld 45 amp dataset all nodes fixed amp sphere_res 64 amp res half As the example shows a view resource request is specified by adding an addi tional parameter resource to what would otherwise be a view request The value of this parameter should be the name of the data vertex which corresponds to the resource to retrieve when the DataLink Server returns branches that correspond to physical resources it inc
105. f the nodes along Mass Ave you can see from Figure 7 7 the vanishing points of the nodes along Mass Ave follow the trajectory of the road as it curves towards the West The Map Viewer uses a simple algorithm for colorizing the vanishing points to make it easier for the user to recognize the common vanishing points of two nodes After the vanishing points are projected to the x y plane of the map viewer they are colored based on the angle of their trajectory In essence the Map Viewer divides the compass circle into 10 degree pie slices each with a different color Which ever slice a vanishing point lies in determines its color just like the direction of the arrow on the spinner of the Twister game determines next color the players have to touch Vanishing points that have near vertical components such as the ones corre 131 Figure 7 8 The node viewers vanishing point visualization sponding to the vertical edges of buildings don t have much meaning in the planar projection of the Map Viewer For this reason the Map Viewer suppresses the display of any vanishing points that have an elevation of more than 45 degrees The vanishing points visualization of the Map Viewer is activated by clicking a check box beneath the map which is labeled vanishing points The vanishing points are visible from the node viewers as well When a user acti vates the vanishing points visualization for a node viewer by selecting the
106. face Over time as the science of geodesy has progressed people have come up with more and more accurate models of the Earth s surface Often these can be used anywhere on the Earth but sometimes they are only meant to be used in specific regions of the globe These models are called datums and consist of an ellipsoidal model of the Earth s surface which is centered at the at the origin of some Cartesian coordinate system The origin of these axes may be located at the true center of the Earth for datums that are meant to be accurate anywhere on the planet or it might be located some distance away in the case of datums which are dedicated to a specific region This type of coordinate system is also sometimes referred to as an East North Up ENU or East North Height ENH coordinate system Hofman Wellenhof et al 1997 5In geodesy almost all models of the Earth s shape are ellipsoidal This is only an approximation the actual shape of the Earth is something more like a distended pear but it is reasonably accurate and makes the mathematics behind geospatial coordinate transformations tractable 107 Datum Reference Ellipsoid GCTL Class generic datum GeoDatum NAD27 Clark 1866 NAD27_Datum NAD83 GRS80 NAD83_Datum WGS84 WGS84 Ellipsoid WGS84_Datum Table 6 1 GCTL Datum Classes The WG5S84 datum which is used by GPS systems and is accurate anywhere on Earth uses an ellipsoid which is almost exactly the same size
107. fine the attributes of the selected resource can lt b gt not lt b gt be null Xk x x x return a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute 8 values contained in lt code gt attrMap lt code gt x since JDK1 0 public static final URL formatViewResourceRequestURL String strNavVtx Name String strData VtxName SimpleNamedVal ueMap attrMap I I II III III II III openResourceStream VITIIITIII III III Requests the resource given by lt code gt branch lt code gt at nav vertex lt code gt strNavVtxName lt code gt from the DataLink Server and returns it as an InputStream ok 154 param branch The Branch which specifies the resource to request param strVtxName Nav vertex of the requested Branch param curVtxAttrsBean Attributes of the currently selected vtx can be obtained from the NavPathBean forwarded by the DataLinkSerlvet return An InputStream for the resource given by lt code gt branch lt code gt at nav vertex lt code gt strNavVtxName lt code gt obtained via the DataLink Server since JDK1 2 public static InputStream openResourceStream Branch branch String strNavVtxName AttributeMapBe
108. for writ ing content handlers These tools include the DataLinkDispatcher which allows a programmer to make requests to the DataLinkServlet for a particular branch or a resource with a given set of attribute values and the BranchHistoryScripter which allows a programmer to insert interactive menus to give the user some sense of the multidimensionality of the data space This chapter also discusses the data structures which are used by content handlers and the DataLinkServlet to exchange information and describes the format of requests made to the DataLinkServlet 4 1 Content Handlers Are JSPs Servlets or HTML Documents Dynamic content handlers can be constructed using Java Server Pages or Java servlets In order to have access to the DataLink Server API content handlers must reference the DataLink Server package by including the line import datalink This gives the content handlers access to the Java classes that are used to com municate with the DataLinkServlet format branches and create context menus as 13 well as the data structures for storing branches and branch classes There are also some data structures for attributes and their values which are located in the package datlink model data attribute There is no reason why HTML documents or anything that can be referenced on the world wide web via a URL cannot be used for content handlers but any components other than servlets and JSPs cannot access the Java based Dat
109. fore to hook up a custom coordinate system to the GCTL library all one has to know is the change of basis transformation from the local coordinates to ECEF If this transformation is not known it can be established using GPS sur veying techniques as described in Hofman Wellenhof et al 1997 or aerophotometry EUROCONTROL and IfEN 1998 Once the change of basis is established it can be written down in homogenous matrix form and used to initialize a LocalGridCoordSys object in GCTL as in the following example which creates a transformation object for for the City Group s LTP coordinate system Mat44 m44CityLTP 0 94603 0 21836 0 23944 1529571 69879 0 32406 0 63746 0 69901 4465216 36854 0 0 73888 0 67383 4275544 15726 0 0 0 1 LocalGridCoordSys cityLTP m44CityLTP Once a local grid coordinate system has been created the LocalGridCoordSys class provides two methods for converting local grid coordinates to ECEF coordinates and back The matrix elements in this example are only shown to five significant digits to save space The real City LTP coordinate system is defined with much greater precision 112 1 Vector3D LGCtoECEF const Vector3D amp v3 const 2 Vector3D ECEFtoLGC const Vector3D amp v3 const 6 4 3 GCTL Can Convert Between Geodetic and Geocentric Coordinates Converting between geodetic and geocentric coordinates of the same datum can be done with an instance of one of the datum cla
110. ftp ftp remotesensing org pub proj PROJ 4 3 12 pdf gt 2 MB Haeberli 2002 Haeberli P 2002 The SGI Image File Format Technical report Silicon Graphics Computer Systems 7 Apr 2002 lt http reality sgi com grafica sgiimage html gt Hofman Wellenhof et al 1997 Hofman Wellenhof B Lichtenegger H and Collins J 1997 Global Positioning System Theory and Practice Springer Vertag New York New York fourth edition Horn 1986 Horn B K P 1986 Robot Vision The MIT Electrical Engineering and Computer Science Series The MIT Press Cambridge Massachusetts Karmien 1999 Karmien R 1999 Music An Appreciation McGraw Hill Higher Education New York New York brief edition Korf 1988 Korf R E 1988 Search A Survey of Recent Results chapter 6 pages 197 237 Exploring Artificial Intelligence Survey Talks From the National Con 182 ferences in ARtificial Intelligence Morgan Kaufmann Publishers San Francisco California Ed Howard E Shrobe McMillan 2000 McMillan L 2000 Computer Graphics 6 837 Course Notes National Geodetic Survey 2002a National Geodetic Survey 2002a NADCON North American Datum Conversion 17 May 2002 lt http www ngs noaa gov TOOLS Nadcon Nadcon html gt National Geodetic Survey 2002b National Geodetic Survey 2002b The NGS Geodetic Toolkit 17 May 2002 lt http www ngs noaa gov products_ services shtml ToolKit gt remotes
111. gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt Cparam strNavVtxName Nav vertex of requested resource param strDataVtxName The name of a data vertex for the type of resource requested Cparam attrMap A set of name value pairings that define the attributes of the selected resource can be null x x return a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt since JDK1 2 public static final URL formatViewResourceRequestURL String strNavVtx Name 153 otring strData VtxName NamedValueMap attrMap Returns a URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt This method is intended for us in applications where Java 1 2 may not be used param strNavVtxName Nav vertex of requested resource param strDataVtxName The name of a data vertex for the type of resource requested param attrMap A set of name value pairings that de
112. he BranchHistoryScripter By now the reader has gotten a sense of how things that are are always in the same place in the fixed data hierarchy can be swapped and moved around in the navigation hierarchy where they may be reachable by more than one path This makes navi gation much more flexible and universal but without some kind of kind of context information to guide the user it can be hard for them to develop a mental model of the data space The BranchHistoryScripter can be used in content handlers to generate navigation context menus that serve as both powerful navigation interfaces and virtual sign posts for the user which help to assuage this difficulty Figure 4 1 shows one of these navigation context menus in action As Figure 4 1 shows the menus that the BranchHistoryScripter produces take the form of a delimited path constructed from the names of the branches that the 83 user has traversed Mousing over one of these branch names brings up a pop up menu which lists all of the other options the user could have chosen in addition to the selection that they did make Selecting on of these options will take the user back to that branch in the navigation hierarchy In Figure 4 1 the user has activated the menu for the pose refinement selection vertex As was mentioned in Section 3 4 1 it is possible for the author of a naviga tion model to suppress history enumeration When enumeration is suppressed the DataLinkServlet will not incl
113. he PIECES vertex looks like PIECES MUSIC_FILES COMPOSER DATASET which given the current attribute val ues expands to a the path classical Brahms music_files Brahms op mp3 where the asterisks denote the unknown values of the pieceName and opusNum at tributes The data layer matches this path to the two Brahms pieces in the dataset and returns the following sets of attribute bindings as the possible branches for the PIECES vertex composer Brahms datasetName classical pieceName Sonata for Clarinet and Piano no 1 opusNum 120 and composer Brahms datasetName classical pieceName Sonata for Clarinet and Piano no 2 opusNum 120 Another way which the navigation model might synchronize with the data model for vertices that do not map directly to data vertex types is illustrated by the green arrow in Figure 2 2 This arrow depicts the opusNum attribute being promoted to a higher level in the navigation hierarchy When the navigation layer explores the opus number vertex it will try to determine all possible values for the promoted opusNum attribute given the current set of attribute values It will do this by consulting with the data layer about the possible values for this attribute in a way that is analogous to requesting all valid branches for the PIECES vertex and extracting from this the set of all possible values for the opusNum attribute each of which will be presented as a distinct navigatio
114. he attribute values given by lt code gt attrMap lt code gt which may be null param strNavVtxName Name of the nav vertex to view param attrMap A set of name value pairings that define the attributes of the selected vertex x ok param pHistory An array containing the names of the n vertices visited thus far which pHistory 0 pointing to the first vertex the root and pHistory n 8 pointing to the latest return a String containing the URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which may be null since JDK1 2 public static String formatViewRequest String strNavVtxName NamedValueMap attrMap Stringl pHistory Returns a String containing the URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained 147 by the attribute values given by lt code gt attrMap lt code gt which must lt b gt not lt b gt be null This method is intended for use in applications where Java 1 2 may not be used param strNavVtxName Name of the nav vertex to view x x param attrMap A set of name value pairings that define the attributes of the selected vertex x return a String containing the URL for a propererly formated request
115. he figure shows the navigation layer can be thought of as a directed graph or network in which each vertex represents a specific location in the navigation hierarchy Leaf vertices shown in Figure 2 1 as darkened circles represent physical resources in the dataset such as images data files etc All other vertices in the navigation graph correspond to locations in the virtual navigation hierarchy and may or may not be associated with locations in the physical data hierarchy A content 30 Presentation Layer Content Handlers JSPs Servlets HTML Applets aL Joes IL 7 BL Data Layer ion Figure 2 1 Interactions between the abstraction layers of the DataLink Server handler in the presentation layer may be assigned by the web architect to each non resource vertex in the navigation network a default content handler is provided for any vertices for which the web architect elects not create a custom content handler Edges in the navigation graph tell the DataLink Server what data is visible to the user at a particular vertex in the navigation hierarchy The two gray arrows in Figure 2 1 show the primary data paths through the DataLink Server Path 1 depicts what happens when a user currently viewing vertex V selects Vz as the next vertex to display The user does this by clicking on some link generated by the content handler for Vi as suggested by the blue arrow going from the presentation l
116. he web interface I ll also document a number of tools which will be useful in the continued development of this project and make some suggestions about powerful additional features that can now be added with ease by taking advantage of the new tools I have created 7 1 Enhanced Visualizations The visualization tools for the pre existing web interface have been re equipped with some new advanced features This section will describe a few of them 121 7 1 1 Extensions to the Map Viewer The development of the Geospatial Coordinate Transformation Library has made it possible to re implement the Map Viewer using a vector format map instead of a crude raster graphic This map is derived from a CAD format base map of MIT s campus which is maintained by the MIT Department of Facilities and is expressed in US State Plane coordinates Using GCTL this map has been exported to the City Group s LTP coordinate system and for the first time the world has a view of how accurately the City group s geospatial imagery is registered with respect to an outside data source The new Map Viewer provides the most accurate depiction of where the nodes in the City dataset are located on campus Figure 7 1 shows a screen shot of the new Map Viewer with the nodes overlaid on the vector format map The numbers displayed in the status window of this screen shot are the coordinates of the mouse s location within the Map Viewer window expressed in
117. ho has experience with one of the markup languages in the SGML family including XML and HTML 3 2 A Block Level Description of the Model File The model file consists of two major sections one which describes the data model and one which describes the navigation model The data model block consists of section in which the types of vertices which can exist in the data model are defined and a 2The parser chosen for the DataLink Server is the Apache Xerces XML parser This parser was chosen because it adheres to the standard XML specifications issued by the World Wide Web Consortium W3C lt http www w3 org gt It is available in both C and Java which makes porting applications to different language environments easier Also it is free and open source Xerces distributions and documentation are available for download from the Apache XML Project website lt http xml apache org gt 3See XML Schema Sperberg McQueen and Thomson 2002 lt http www w3 org XML Schema dev gt See The World Wide Web Consortium 2002a 46 section which describes how instances of these vertex types are connected together to construct a model of the physical data The navigation model block contains definitions of all the navigation vertices in the navigation model and a specification of how all of these vertices are connected A skeleton of the model file looks like lt xml version 1 0 encoding UTF 8 gt lt modelDefs xmlns xsi http
118. ife cycles of servlets and Java Beans 76 Method Return Value getCurVtxAttrs An AttributeMapBean containing the attributes bindings of the current navigation path getHistory A String containing the current navigation history retrieveBranchClass The BranchClass for nav vertex nav vertex branchClasses An Iterator containing the names of all the children of the selected vertex for which the DataLinkServlet has found branches Table 4 1 NavPathBean routines for use in content handlers and the available navigation options Table 4 1 summarizes these methods As you can see from Table 4 1 the curPathBean uses a number of additional data structures for encapsulating the information it holds about branches and attributes 4 3 2 Data Structures for Attributes Table 4 2 lists the three map classes that are useful in content handlers for storing sets of attribute name value pairs All of these maps have the same basic inter face which includes a get name method for retrieving a named attribute value and a number of put name V methods for storing the value of an attribute NamedValueMap and SimpleNamedValue map are essentially the same except that SimpleNamedValueMap has a simplified API that makes it compatible with Java 1 0 SimpleNamedValueMap is the storage class of choice for use in Java applets embedded in content handlers because its JDK 1 0 compatibility will ensure that the applet can run in
119. igation model can override the default values of these parameters for some vertices by including a param command in the body of the navigation vertex s definition The param command has the syntax lt param name parameter_name value true false gt 65 supress_incomplete_leaves The supress_incomplete_leaves parameter is relev ent to the opus number navigation vertex that has just been added to the navigation model As it stands now when the user visits this vertex from the classical lib vertex the DataLink Server will display branches for every composer in the dataset that has written a piece with the selected opus number as well as a set of branches for every audio file in the dataset with the selected opus number This might be the behavior intended by the web architect On the other hand if the purpose of the opus number vertex was merely to reverse the order in which the user can select values for the opus number and composer name attributes then the DataLink Server shouldn t display audio files when the user is at the opus number vertex unless the user has arrived there from the composer vertex and thus has selected a value for the composer name attribute To force the DataLink Server to take this second approach the suppress_incomplete_leaves parameter should be set to true for the opus number vertex as follows lt vertex name opus number gt lt child name composer gt lt child name pieces gt lt deriv
120. information about the node including images links to the pose and feature data as well as some other visualization tools Clicking on an edge between two nodes brings up the Epipolar Viewer see below for the implicated node pair Figure 1 1 is a screen shot of the Map Viewer in mini mode The Mosaic Viewer This tool allows the user to view the spherical texture from the point of view of the optical center of the node looking out Using the mouse or the keyboard the vantage point can be rotated vertically or horizontally and the field of view can be zoomed in or out The Mosaic Viewer in the pre existing inter face is equipped with a gamma correction tool which compensates for the sometimes strange coloration of the images due to the fact that the images have a high dynamic range logarithmically mapped onto a linear scale from 0 255 by boosting the color saturation of the image Images in the city dataset are encoded using the SGI rgb file format Haeberli 2002 19 The rgb format is not very web friendly it is neither supported by most web browsers nor the standard Java API To ameliorate this problem the existing toolkit used to rely on a script that was run prior to the deployment of the web interface which generates jpg encoded versions of the spherical textures One of the extensions to the pre existing web interface was to augment the Mosaic Viewer and the other applets that use rgb images with a Java utility class RGBImage
121. init sizeof pParams sizeof char pParams gt I m not taking sides on an OOP vs no OOP debate here I m just making an observation about an example where OOP is better suited to the problem 103 fprintf stderr Couldn t initialize state plane projection projUV uv UV uv uv u V T3 65 pj fwd uv pStatePlaneProj printf x f y f uv u uv v This is not egregious but it s not superb either There s no need to spend time lambasting it but a couple of observations about this sample will demonstrate the advantages of GCTL The generic initialization function necessitates an awkward calling convention More importantly however without precise arguments there is no way a pro grammer can even begin to setup this transformation without diving into the documentation There is no way to tell which one of uv u and uv v is latitude and which is longitude The universal coordinate data type uv which is both the input and output of pj fwd makes it impossible to tell what type of coordinates the data is really expressed in If the call to pj fwd was isolated from the initialization and output code as it is likely to be in a real program it would be impossible to infer whether pj fwd is transforming to or from state plane coordinates Now consider the same transform implemented using GCTL Initialize a State Plane projection for Mass mainland zone 2001
122. ion are approximations The method for transforming between geodetic and geocentric coordinates that was used to implement the geocentric to geodetic transformation in the GeoDatum class is the Toms method Toms 1996 Toms method is relatively new and it hasn t appeared in much of the literature yet The most common method currently in use is Helmert s formula EUROCONTROL and IfEN 1998 which exists in both an iterative and an analytic form but this method is much less accurate than Toms In testing on the City dataset and a few other data sources both versions of Helmert s transformation were only able to produce results to within a few meters accuracy Toms method on the other hand was able to produce results with centimeter accuracy on the same data It takes the following steps 1 R Vr y Compute the distance from the z axis 2 vo Kz Estimate distance of 9 0 from the equatorial plane 3 ho yo R2 Estimate radius of 0 0 4 v z be 72 Revise estimate of vo 5 S R ae Estimate the radius of 0 0 projected onto the equatorial plane 11 Theoretically Helmert s method can produce results with centimeter accuracy but small errors in the source data tend to blow up fairly quickly especially when the ellipsoid parameters for the reference datum are imprecise 114 6 hy v S Revise estimate of ho 7 sing 7 Compute sin of latitude 8 cosd Comp
123. ith the attribute values contained in lt code gt attrMap lt code gt x since JDK1 2 public static final InputStream openResourceStream String strNavVtx Name NamedValueMap attrMap throws IOException x Requests the resource at nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt from the DataLink Server and returns it as an InputStream This method is intended for use in applications where Java 1 2 may not be used param strNavVtxName Nav vertex of requested resource attributes of the selected resource can lt b gt not lt b gt be null ok x param attrMap A set of name value pairings that define the return a String containing the URL for a properly formatted request to the DataLinkServlet for the resource corresponding to lt code gt strDataVtxName lt code gt viewed from nav vertex lt code gt strNavVtxName lt code gt with the attribute values contained in lt code gt attrMap lt code gt since JDK1 0 public static final InputStream openResourceStream String strNavVtx Name SimpleNamedValue Map attrMap throws IOException LITTLTTTLTT TTS sortBranches 107 I I II III Sorts the branch array lt code gt pBranches lt code gt lexicographically by branch name using the quick sort algorithm param pBranches Array of branches to sort public sta
124. iven by http city lcs mit edu 8080 data2 DataLink The name of the DataLinkServlet has been shortened to DataLink in this example It is possible using configuration parameters of the webapp which deploys the DataLink Server to re map the URL to the DataLinkServlet to anything that is convenient As Section 4 2 indicated the view parameter of the DataLink Server request specifies the name of the next navigation vertex to display That happens to be the node vertex in this example The history parameter is a delimited list of the navigation vertices visited thus far along the current navigation path The vertices visited are listed in order from left to right beginning with the first vertex visited on the left As was mentioned in Section 4 2 1 the history is passed in the DataLink Server request in order to maintain a stateless architecture for the DataLinkServlet When necessary it is the joint responsibility of the DataLinkServlet and content handlers to ensure the the history is updated properly When it is not relevant the history parameter can be omitted from view requests The rest of the parameters of this DataLink Server request example are attribute bindings In this case the request specifies that the value of the dataset attribute 89 should be all nodes fixed and the value of the nodeId attribute should be 75 This request is asking the DataLink Server to return all of the branches for each of the children o
125. ld Wide Web Constortium 11 May 2002 lt http www w3 org gt Toms 1996 Toms R 1996 An Improved Algorithm for Geocentric to Geodetic Coordinate Conversion Technical report Warmerdam 2000 Warmerdam F 2000 Proj 4 Cartographic Projections Library Technical report remote sensing org 14 Mar 2001 lt http www remotesensing org proj gt 184
126. lection of geospatial imagery on the world wide web The two driving visions behind this project were 1 to develop a system for automatically providing flexible modular and customizable infrastructure for navigating a hierarchical dataset based on a simple and concise description of the paths users can take to reach items in the dataset and 2 to create a visually exciting interactive medium for members of the machine vision community to explore the data in the City Scanning Project Dataset and learn more about techniques the City Scanning Group is developing for rapidly acquiring geospatial imagery with very high precision In an effort to realize these goals I have developed the DataLink Server a system that provides navigation and data retrieval infrastructure for a dataset based on a specification written in a descriptive and coherent XML based web navigation and data modeling language a geospatial coordinate transformation library for converting coordinates in virtually any geographic or cartographic coordinate system and a battery of visualization applications for exploring the epipolar geometry baselines vanishing points edge and point features and position and orientation information of a collection of geospatial imagery These tools have been successfully used to deploy a web interface to the more than 11 000 geospatial images that make up the City Scanning Project Dataset collected by the MIT Computer Graphics Group Thesis Supervisor
127. ll gamut of format tokens that C programmers may know about there was no need for floating point attributes in the City dataset but the string format expression code is modular enough that format tokens for new data types can be added without tribulation Other than the floating point tokens most of the other C format tokens can be subsumed by the d token anyway In addition there is no support for the hexidecimal format modifier and I do not foresee there ever being a need for the modifier 54 int hd string AS stateVar abstract state stateVar literal state Table 3 3 String format expression tokens used by the DataLink Server DataModel lt name DATASET type dataset gt dafsetName lt name COMPOSER type composer_dir gt composerName lt name MUSIC_FILES lt name BO DIR type music_dir gt type bio dir gt lt name PIECES lt name BIOGRAPHY lt NGMe PORTAI type mp3 gt type bio_doc gt type portalt_img gt opusNum pleceName Figure 3 1 Data model of a classical music library will give preference to the data vertex type signature which appears first in the body of a data vertex type definition 3 3 2 Declaring Data Vertex Instances Now that that the tools for describing the data have been garnered it s time to learn how to declare instances of these data types and connect them together to form the data model graph An example is the be
128. lt navModel gt In this case when the user chooses a state of post processing refinement first the system should not behave as if a value for the refinement attribute has also been selected The DataLink Server recognizes this case when it finds two branches with identical names but different sets of attribute bindings For example if the user made a selection for a state of post processing refinement before choosing a node ID the system would find a whole bunch of branches named half for the image resolution attribute The DataLink Server knows better than to provide selections for each of these branches The DataLink Server would recognize that some of these like named branches have different values for the node ID attribute triggering the branch merging process which unifies all of the branches with similar names into one 67 When two branches are merged they are generalized This means that all of the attribute bindings for attributes that were not set by the user via navigation which do not belong to the data vertex type if any of the navigation vertex of the branches are annulled Thus if the DataLink Server identified these two branches for the resolution vertex nodeId 0 refinement rotated resolution half and nodeId 1 refinement rotated resolution half when the user had just made a selection for refinement the DataLink Server would forward only one branch to the presentation layer This branch woul
129. lt TD gt lt INPUT TYPE submit NAME action VALUE Stop Server gt lt TD gt pw printin lt TR gt pw printin lt TABLE gt pw println lt FORM gt 179 pw printin lt BODY pw printin lt HTML gt end doPost IIIIII destroy IIIIIII Called when servlet is being taken out of server This method x ensures the QutputStreamServer is shutdown at this time public void destroy synchronized m_serverMutex if m_OSServer null System out printin Monitor servlet is being taken out of service m_OSServer stop System setOut m_oldSysOutStream end destroy end MonitorServlet 180 Bibliography Antone and Teller 2000 Antone M E and Teller S 2000 Automatic Recovery of Relative Camera Rotations for Urban Scenes In Proc Proceedings of CVPR 2000 Volume 2 pages 282 289 Antone and Teller 2001 Antone M E and Teller S 2001 Scaleable Absolute Position Recovery for Omni Directional Image Networks In Proc Proceedings of CVPR 2001 Bodnar 2001 Bodnar Z M 2001 A Web Interface for a Large Calibrated Image Dataset Advanced undergraduate project MIT MIT Computer Graph ics Group 200 Technology Square rm NE43 253 Cambridge Massachusetts lt http graphics 1lcs mit edu seth pubs gt Bosse et al 2000 Bosse M De Couto D S J and Teller S 2000 Eyes of argus Georeferenced imagery
130. ludes this information in the curPathBean for use by the DataLinkDispatcher 5 2 3 Specifying Abstract State Values When making a request to the DataLink Server it is often necessary to make requests with state attribute values given as abstract states After all the whole idea of 90 the DataLink Server is to abstract away the hierarchy of the physical data so the navigation and presentation layers should not have to know the literal state values of state attributes are To specify the value of a state attribute as an abstract state instead of a literal state the attribute binding in the request should take the form of this example refinement initial Here initial is the name of the initial abstract state of the pose_refinement attribute prefixed with a character to indicate that the value is an abstract state instead of a literal state To specify a literal state the client lists the attribute binding with the name of the literal state omitting the prefix 5 3 How the DataLink Server Handles View Re quests The best way to describe how the DataLink Server handles a view request is to walk through one step by step This section will do just that beginning with what the DataLink Server does as soon as it receives a request It may help to have a visual reference to accompany this discussion so the data paths diagram from Chapter 2 is repeated here in Figure 5 1 for convenience The first thing the DataLink S
131. making a conscious effort not to make this read too much like the liner notes of a CD T d like to briefly mention The Talking Heads Nirvana The Cure The Spin Doctors Counting Crows Train Brahms Mozart Beethoven Gershwin Copland Isabelle Allende Gabriel Garcia Marquez Jim and William Reid Jake Burton and the seminal noise acoustics band Sonic Youth And now for the real funky stuff Contents 1 Introduction and Background Information 15 1 1 Background Information 0024 15 1 1 1 Image Acquisition and Post processing Stages 16 1 1 2 Description of the Existing Interface 18 1 2 Overview of the Extended Interface 23 1 2 1 The Datalink Server 24424 4 nee 23 1 2 2 The Geospatial Coordinate Transformation Library 25 1 2 3 Other Minor Extensions and Improvements 27 2 The DataLink Server 29 2 1 The DataLink Server Maps Virtual Navigation Hierarchies to Real Data 30 2 1 1 The DataLink Server Can be Accessed Programatically 33 2 2 Components of the DataLink Server 33 2 2 1 Data and Navigation Models Define the Abstraction 34 2 2 2 The Core of the DataLink Server is the DataLinkServlet 39 2 2 3 The Front End Consists of Content Handlers 39 3 Modeling Data and Directing Navigation Using XML 45 3 1 The Model Definitions are Described in XML 45 3 2 A Block Level Description of the Model
132. mply a C style string format expression with the token 4d in the place of the image ID number The signature expression indicates that the value of this token is equal to the value of the imgId integer attribute String format expressions like these are all that is necessary to extract the salient attribute values from file and directory names in a directory hierarchy Although it was not necessary for the City Scanning Project Dataset a similar mechanism could be added to the data model for extracting attribute values from data actually stored within the data files themselves and it would not be much of a stretch to add a parallel construct for obtaining attribute values from database keys Table 3 3 lists the tokens that can be used in signature expressions Most of these are identical to their counterparts in C In addition I have added two new tokens t and T for formatting state attributes It is not uncommon to find datasets with inconsistent naming schemes To combat this problem the DataLink Server supports the capability to define multiple signa tures per data vertex type As is the case for state signatures the DataLink Server 6 This discussion assumes the reader is familiar with the syntax of C string format expressions However for a reference or a quick review look in any C C guidebook I recommend C The Complete Reference Schildt 1995 The current implementation of the DataLink Server does not support the fu
133. ms together 109 As Section 6 4 4 will describe GCTL performs datum shifts by performing a change of basis on geocentric coordinates This matrix has the form o Fe ry AA r oO fa AY ly AZ 6 1 AX AY and AZ are the translational offsets from center of the new datum s ellipsoid to the origin of WGS84 and correspond to deltaX deltaY and deltaZ respectively The scale parameter specifies o the scaling factor for changing units between the two datums The last three parameters rotX rotY and rotZ correspond to Tz Ty and rz respectively These parameters describe the rotation between the new datum s axes and WGS84 s These parameters are a fairly standard way of defining datum shifts and can usually be looked up in a table somewhere such as EUROCONTROL and IfEN 1998 To define a new ellipsoid for use in GCTL one can derive a new ellipsoid class from the generic Ellipsoid class as follows class New_Ellipsoid public Ellipsoid public New_Ellipsoid Ellipsoid majorAxis flattening F In this example majorAxis is the semi major axis of the ellipsoid flattening is the geometric flattening which is a special parameter that is commonly used to describe geodedic ellipsoids and is defined by 6 2 where a is the length of the semi major axis of the ellipsoid and 6 is the length of the semi minor axis Hofman Wellenhof et al 1997 110 6 4 Using GCTL In this section the rea
134. n option to the user The Models are Defined Using XML The models for a particular dataset are defined in an XML document which is com pletely separate from the rest of the DataLink Server The model definition file The details about how the DataLink Server maps attribute values to directory and file names will be given in Chapter 3 38 functions as a configuration file for the DataLink Server which is used to bootstrap it with the abstraction function for a particular dataset so configuring the DataLink Server for a new dataset is chiefly a matter of writing a model file for the new dataset In addition modifying the layout of an existing site is as simple as altering a few ver tex definitions in the model file The beauty of this mechanism is that alterations to the layout of web interface don t require rewriting the hyperlinks HTML documents and scripts that make up the front end of the interface in order to re wire the website because DataLink Server follows the specification in the model file to do all of this automatically Also data can be added to or removed from the dataset without the need to regenerate any part of the web site s infrastructure since all of the navigation options are determined dynamically The content handlers to be used for each nav igation vertex are also specified within the model file Chapter 3 will have more to say about the navigation and data model abstraction and the XML based modeling
135. nch named after c with the attribute bindings Be to the curPathBean and go back to Step 4 to process the next child 4 4 2 Otherwise promote the attribute a associated with c s derived prop erty by adding a null valued binding for a to B and setting a flag bDerivedProp which indicates that all branches for c with the same value for a should be merged and generalized Set T to the source data vertex type of the derived property 4 5 If c has any properties insert them in B overriding any conflicting at tribute bindings 4 6 If c redirects any attributes rename the appropriate bindings in B 4 7 Make a request to the DataModel for all of the sets of attribute bindings which map to data of type T and are consistent with B 4 8 Unless the merge_duplicates flag is set to false for this vertex check to see if c is not a leaf vertex or if the flag bDerivedProp is set and if 93 either of these conditions is true merge and generalize all of the duplicate branches returned by the DataModel if any 4 9 Add the branches if any to the curPathBean and loop 5 If the suppress_incomplete_leaves flag is set for vy then remove any branches from the curPathBean for leaf vertices with which are not completely specified by the bindings in Bin 6 Append vy to the history 7 Forward the curPathBean to the content handler for vy This is a lot of steps to be sure but most of them are pretty straightforward On
136. nes whether the DataLink Server should take Path 1 or Path 2 If the answer is Path 1 the NavModel will have to make requests to the DataModel for the branches of each of the children of vy If it is Path 2 the NavModel will have to ask the DataModel to resolve the path to the appropriate resource and serve it to the client Here is an enumeration of the steps the NavModel takes in this process 1 If vy has any properties associated with it named value pairs for each of these are added to Bin If any of the properties are in conflict with attribute bindings in Bin these attribute bindings are replaced by the property values 92 2 If there are any attributes in B which are redirected by un these attributes are renamed at this time 3 If vy is a leaf vertex the NavModel makes a request to the DataModel to resolve the path to the corresponding resource given the attribute bindings in Bin and serve the data to the client Otherwise the NavModel proceeds to Step 4 4 Loop over each of the children of vy 4 1 Get the next child c of vy 4 2 Check to see whether c is in the history If it is skip c and go back to Step 4 to get the next child vertex 4 3 Create a duplicate of Bin Be 4 4 Check whether c is associated with a data vertex type T If so jump ahead to Step 4 5 Otherwise c is a purely navigational vertex 4 4 1 If there are no derived properties associated with c add a single bra
137. ng all of the node location data into one file and then serving only this file to the applications that request location information for a large number of nodes in the dataset The applications are then responsible for parsing the node locations file to extract the data that they need Caching can be implemented fairly easily using the DataLink Server For node locations it is done by creating a special node locations navigation vertex The con tent handler for this vertex is simply a special servlet which serves the cached node locations file to the client application when this file is up to date When the cache file is not up to date this servlet is responsible for regenerating it which is done by making recursive requests to the DataLink Server to retrieve and open the node 0 4 lt http city 1lcs mit edu 8080 data2 open_source rgb gt 136 camera files for each node To implement a caching content handler like this is not difficult The only require ment in terms of the DataLink Server configuration is that the navigation vertex to which the caching content handler is affixed must have a child vertex for data files which are used to generate the cached data The caching content handler simply checks the time stamps of each of the branches of this vertex when it recieves a request to see whether the cache is up to date The way that the caching content handler determines whether the cache is up to date with the data is by first check
138. nition of the model file format When the model file is parsed this schema is used to check and report errors in the model file s structure and syntax To ensure that errors are properly reported during parsing a reference to this schema should be included in every model file A complete listing of model file schema is located in Appendix B 3 2 2 Reserved Characters In general vertex and attribute names in the model file can contain any character including spaces Some of the special characters used in requests to the DataLink Server will cause problems if they are used in attribute or vertex names Table 3 1 lists these characters and summarizes the contexts in which they are reserved 3 3 Creating a Data Model This section describes how to model a hierarchical dataset The example datasets in this document will be directory hierarchies because the current implementation of the DataModel class is geared towards modeling directiories This is a reflection of the first target dataset of the DataLink Server the City Scanning Project Dataset which is kept on disk in an elaborate directory structure It should be noted however that the data model is merely an abstraction and that although the software implementing 48 this abstraction is tailored to modeling directory trees it would be easy to extend the data model to provide an interface to any hierarchical data storage system such as a database With that in mind it
139. nk Server will display a listing for the classical lib vertex that looks something like Figure 3 3 Clicking on one of the composers names will take the user to the composer vertex where the DataLink Server will display something similar to Figure 3 3 with sections for pieces portrait and bio Observe how the child command is used to create these edges in the navigation graph by naming the next vertex a user could be able to get to from a given location At this point the reader may have noticed something queer about navigation 60 Current Path classical lib Antonio Vivaldi Franz Schubert 8 8 George Gershwin EEE Johannes Brahms Ludwig Wan Beethoven Modest Mussorgsky Wolfgang Amadaeus Mozart Figure 3 3 Listing of the classical lib vertex model and the output of the DataLink Server for the classical lib vertex At no point in the navigation graph was the user given an option to choose the value of the datasetName attribute which according to the data model that was developed in Section 3 3 2 is necessary to resolve a path to the composer directories Somehow the DataLink Server was able to produce results for the composer vertex anyway To understand how this happens recall the specification of the navigation layer s interaction with the data layer When the navigation layer requests a set of branches for a data vertex type it is asking for all sets of attribute bindings which completely specify a data and are consistent
140. nt to show how the well the rotation phase accomplishes this task by demonstrating how closely the vanishing points between two nodes are aligned relative to each other and how accurately the vanishing points line up with parallel lines in the real world Figure 7 7 shows a zoomed in view of the vanishing points of the nodes along Massachusetts Ave near MIT The vanishing points are drawn as line segments with their midpoints at the location of the node in which they were detected The length of these line segments from the midpoint to one endpoint is the distance between the source node and the source node s furthest adjacent neighbor This allows the alignment of the vanishing points between two adjacent nodes to be compared If the two nodes are well registered rotationally the vanishing points in the map viewer should appear to be parallel Overlaying the vanishing points on the map of campus allows the user to see how they correspond to the physical geometry of the world As Throughout this discussion P ve used the perspective transformation for an ideal pinhole camera with unit focal length This is valid for images in the City dataset because they are undistorted before entering post processing in order to remove the radial distortion that is introduced by the real camera s lens making the pinhole projection mathematics applicable 130 a Poe ll XA Bere 5 R ad VA au Figure 7 7 Vanishing points visualization o
141. nto classical forms He was born in Hamburg Germany when his father made a precarious living as a bass player At thirteen Brahms led a double life during the day he studied piano music theory and composition at night he played dance music for prostitutes and their chents in water front bars On his first concert tour when he was twenty Brahms met Robert Schumann and Schumann s wife Clara who were to shape the course of his artistic and personal life The Schumanns listened enthusiastically to Brahm s music and Roberts published an article hailing young Brahms as a musical messiah As Brahms was preparing new works for an eager publisher Schumann had a nervous collapse and tried to drown himself When Schumann was committed to an asylum leaving Clara with seven children to support Brahms came to live in the Schumann home He stayed for two years helping to care for the children when Clara was on tour and becoming increasingly involved with Figure 2 4 How the content handler for the composer vertex might format the data for a composer 41 GV This repository is still under development We would appreciate your feedback Current Path all nodes fixed Dataset Node Node 34 choose another N Node34 Acquired on May 18 2000 2 59 PM Node Viewed ak a Oylindeical Mosaic Node s Position and Nearest Bu hiis planar projection EN mosaic Neighbors rotated translated Show baselines 7
142. odes in the City dataset which is why the context navigation menu for nodes only lists two options one for the currently selected node and one which warps the user back to the next higher level in the hierarchy the dataset vertex in order to choose another When there are only a few alternatives the menus can display them all Parameters in the model file allow the developer of the web site to specify which of these alternatives is appropriate for a given navigation vertex As is to be expected there are performance issues associated with allowing the full listing capability for vertices which could potentially have a large number of branches 43 44 Chapter 3 Modeling Data and Directing Navigation Using XML In this chapter you will learn about the XML based modeling language for describing a dataset and controlling the behavior of the navigation layer of the DataLink Server You will learn how a data model describes how to extract the salient properties of the data from a hierarchical dataset as well as how the navigation layer uses the attributes of the data as the pivots about which navigation is guided You will see how the navigation model outlines the behavior of the navigation layer and how it determines the type of information that is passed to the presentation layer Each of the modeling constructs in this chapter is illustrated with concrete pedagogical examples or real samples from the City dataset and by the end of it you will
143. of attribute value The function of the history is clearly a pretty important one which might leave the reader asking why it is passed around as a parameter in DataLinkServlet request and not maintained somewhere in the internal state of the DataLink Server or in a Java Bean The reason is that it is important for the DataLink Server to maintain a stateless architecture because it is implemented as a Java servlet Well designed Java servlets should not maintain any state because the length of their lifecycle is not guaranteed The servlet container can start and stop servlets at will as long as they are not currently handling a request in an effort maintain a good load balance on the host machine This means that subsequent requests to the DataLink Server may be handled by different run time instances of the DataLinkServlet class and conse quently it is dangerous to try and maintain an internal state in the DataLinkServlet For a similar reason it is also not a good idea to maintain state information like the 19 navigation history within Java Beans Another reason that the history is passed as an argument is that it serves as an im plicit request for history enumeration which can be used to make context navigation menus like the ones that are created by the BranchHistoryScripter Section 4 5 in this chapter will expound the details of this use of the history further It is not necessary for the content handler developer to understand th
144. of the web layout and a similar XML based model of the physical data the DataLink Server automatically resolves the mapping between the logical hierarchy of the pre sentation layer and the location of the physical data on they fly as the information is browsed by a user In addition the DataLink Server allows the web architect to specify content managers which are additional JSPs or servlets that generate the HTML for presenting a particular datum giving the web designer complete control over the look and feel of the front end This infrastructure provides a number of key benefits over the pre existing script generated architecture e As new data is added to the dataset it becomes instantly available for browsing via the web interface because the mapping between the web layout and the physical layout is resolved at run time e No broken links result from the renaming or removal of data because the DataLink Server will only resolve mappings to existing physical data e Information about the layout of the presentation layer is centralized in a single XML document so reorganizing the entire web site can be done by editing a single file 24 e The description of the physical data hierarchy is described in a solitary XML file as well so restructuring of the physical data hierarchy only necessitates modifications to one document in order to update the web interface e The DataLink server allows the web architect to describe multipl
145. older browsers The two named value map classes are utility classes for passing arguments to the DataLinkDispatcher to form requests to the DataLink Server They should be used whenever the content handler needs to make a request using attribute values that were not provided as part of the current navigation path or a branch by the DataLinkServlet If the navigation model for the dataset is designed properly there should almost never be a need to do this in the body of a content handler because the 17 Values Type Usage AttributeMapBean AttributeBean Used to store attributes bindings pro vided by the DataLinkServlet NamedValueMap String Used to form custom queries JDK 1 2 SimpleNamedValueMap String Used to form custom queries JDK 1 0 Table 4 2 Map data structures for holding attribute sets web architect can usually supply all of the data a content handler will need by making edges to the appropriate navigation vertices The content handler can then access the data using the routines from the DataLinkDispatcher interface that can operate directly on the branches produced for these vertices The two primary exceptions occur when the value of an attribute comes from a cache or when the request is made from within a Java applet which cannot have direct access to the curPathBean The AttributeMapBean is the representation used by the curPathBean to store attribute bindings forwarded by the DataLinkServlet Attributes are stored in an
146. ollow Geodetic Coordinate Systems are geospatial coordinates expressed in terms of lattitude longitude and altitude with respect to a specific model of the Earth called a datum Geocentric Coordinate Systems are geospatial coordinates expressed with re spect to a set of Cartesian coordinate axes The origin of the geocentric co ordinate axes is the center of the ellipsoidal approximation of the Earth s shape that is used by the particular datum in which the geocentric coordinates are expressed 106 Earth Centered Earth Fixed ECEF Coordinates is a special name for geo centric coordinates expressed relative to the WGS84 datum Local Grid Coordinates are Cartesian coordinates associated with an arbitrary set of coordinate axes Typically the coordinate axes are defined in terms of a local tangent plane LTP system Local Tangent Plane coordinate systems consist of a right handed set of coordi nate axes with the z axis normal to the Earth s surface and the x and y coordi nate axes lying in a plane tangential to the Earth where the x axis points East and the y axis points North Cartographic Coordinate Systems are 2D projections of geospatial coordinates that are used in maps Unless otherwise noted geocentric ECEF and local grid coordinates are all ex pressed in units of meters Different cartographic coordinate systems may use US feet meters or other units 6 3 1 Datums are Models of the Earth s Sur
147. or class also provides a routine for doing inverse cartographic projections LLA toLLA const Vector3D amp v3Proj const The results returned by this method will always have a zero altitude however since there is no height information stored in the 2D cartographic coordinates This does not however result in a loss of accuracy in the other two dimensions 6 5 Chapter Summary Upon finishing this chapter the reader should have a pretty good idea of GCTL s capabilites as well as how to extend them While mastery of the intracacies of geospatial coordinate transformation comes with practice this chapter has provided the principle details that the reader should need to get started One of the many benefits of GCTL extolled in this text is the way that it makes coordinate types and reference frames explicit It is my hope that this dramatically accelerates the learning curve of any users who begin incorporating this transformation library in their own programs 120 Chapter 7 Advanced Features of the Extended Web Interface The addition of the new navigation data retrieval infrastructure provided by the DataLink Server and the geospatial coordinate transformations afforded by GCTL have made possible a wealth of new features for the extended web interface to the City Scanning Project Dataset This chapter serves as a showcase of this new technology in which I will describe a number of extensions to the visualization tools in t
148. ors a compass and wheel odometers which are used to obtain reasonably accurate position and orientation estimates for each of the images Bosse et al 2000 When the sensor is activated at a given location the camera head rotates about a fixed optical center capturing between 20 and 71 images tiling a portion of the sphere One such collection of images in combination with the associated position and orientation information is referred to as a node The images for each node are numbered sequentially beginning at zero The posi tion and orientation data for each image is logged in a camera file with an ID number matching that of the associated image The camera files generated by the sensor are what is referred to as the inztial pose data When the sensor has been returned to the lab and the new nodes have been added to the City database an edge and point feature detection algorithm is applied to the images Then the initial pose files are subjected to a series of post processing algo rithms to refine the position and orientation estimates All of these algorithms are applied in a sequence of four post processing stages feature detection mosaicing ro lFor more information about the format of a camera file and its contents please refer to Cali brated Registered Images of an Extended Urban Area Teller et al 2001 16 tational alignment and translational registration geo referencing Each of these stages outputs a new set of camera
149. ource destination name of destination gt One example of where this is useful is a pair of special navigation vertices which bring up the baseline files for a particular node in the City dataset Baseline files have a signature which looks like 64 Parameter Default Value supress_incomplete_leaves false merge_duplicates enumerate_history Table 3 4 Default values of navigation vertex parameters lt signature gt 4d_ 4d trans fromNode toNode lt signature gt Attribute redirection can be used to find the to baselines for the currently selected vertex The to baselines are the baselines which point from the current node to another node i e the baselines for which fromNode is the currently selected node To find the to baselines for the currently selected vertex the City dataset employs a special navigation vertex to baselines which has the definition lt vertex name to baselines type baseline_file gt lt redirect source nodeld destination fromNode gt lt vertex gt When the datalink server encounters this vertex it assigns the value of nodeld if any to fromNode A similar vertex is used to find the from baselines of a node Navigation Vertex Parameters Navigation vertices have a small number of Boolean parameters which affect how they are handled by the DataLink Server The names and the default values of these parameters are summarized in Table 3 4 The author of a nav
150. particular node in the City dataset Navigation vertices however need not correspond to a real physical datum at all They may simply exist for the purpose of guiding the user s navigation by assigning values to attributes or they may be used for generating a particular view of some subset of the data The navigation model is coupled with the data model to resolve the mapping between the virtual data hierarchies and the real physical data Vertices in the data model correspond to items in the dataset Edges in the data model graph describe the hierarchical relationships between these elements In particular if there is an edge from data vertex Vi to data vertex V2 then V inherits the attributes of Vo Every vertex in the data model is an instance of a data vertex type Each unique kind of datum in the dataset is represented by its own data vertex type whose definition contains information about the names and types of attributes associated with that particular datum There may be multiple instances of a single data vertex type in the data model The illustration in Figure 2 2 should help to clarify how the navigation model and data model abstractions work together Figure 2 2 shows a simple set of models that describe the web interface to an imaginary classical music library On disk the phys ical hierarchy of the library consists of a root directory which holds a subdirectory for 34 DataModel Navigation Model lt name DATASET lt n
151. pe mp3 gt lt vertex name portait type portait_img gt lt vertex name bio type bio_doc gt lt navModel gt 99 This exposes the images biography text documents and audio files to the naviga tion layer by assigning navigation vertices to them The type argument grounds each of these vertices to the appropriate data vertex type in the data model When the navigation layer explores a navigation vertex with a type declaration it knows to ask the data layer for instances of that data type As it stands however this navigation model doesn t provide any way of reaching this data because no navigation paths have been specified yet The following code listing adds a simple one lt navModel gt lt vertex name classical lib gt lt child name composer gt lt vertex gt lt vertex name composer type composer_dir gt lt child name pieces gt lt child name portrait gt lt child name bio gt lt vertex gt lt vertex name pieces type mp3 gt lt vertex name portait type portait_img gt lt vertex name bio type bio_doc gt lt root name classical lib gt lt navModel gt The path described here begins at the vertex named classical lib as spec ified by the root command From here the user will be allowed to select a com poser directory to view The DataLink Server will present a composer link for every composer_dir it finds in the dataset The default content handler of the DataLi
152. ransformation APIs available stick to basic C imple mentations in an effort to make them simple portable and generic This turns out to be a pain more than a panacea however because the typical API under this model consists of a single general initialization routine that sets up an individual transformation and another solitary generic transformation routine This is problematic because it necessitates the use of global static variables forcing the programmer to reinitialize the transformation methods every time they want to switch transforms Almost all coordinate transforms require so many static parameters that is much better from an architectural point of view to encapsulate the transforms in powerful classes tailored to specific coordinate frames 6 2 GCTL Makes Reference Frames Explicit In addition to being a universal coordinate transformation tool the design of GCTL s API is geared to tackling problems 2 and 3 as well Consider the following example for projecting latitude and longitude coordinates to US State Plane Coordinates in the Massachusetts mainland zone using the Proj 4 library Warmerdam 2000 which is typical of most publicly available coordinate transformation APIs pParams 0 new char 12 pParams 1 new char 16 Initialize a State Plane projection for Mass mainland zone 2001 pParams 0 units us ft pParams 1 init nad27 2001 PJ pStatePlaneProj if pStatePlaneProj pj_
153. rdinate systems 143 144 Appendix A The DataLinkDispatcher API I II II III I II 11 setDataLinkURL 111111111111111 Sets the url to the data link server relative to the current webapp param strDataLinkServletURL url to the data link servlet public static void setDataLinkURL String strDataLinkServletURL IIIII III III III getDataLinkServer LILITTTTLTTTTTT TT TT Gets a String representation of the url to the data link server relative to the current webapp return a String representation of the url to the data link server relative to the current webapp public static String getDataLinkServer III III III III formatViewRequest 145 x x 111111111111111 Returns a String containing the URL for a propererly formated request to view the branch lt code gt branch lt code gt of nav vertex lt code gt strNavVtxName lt code gt The caller should ensure that the history of vertices that have been visited so far are passed to the method using the lt code gt pHistory lt code gt argument if the context requires that current navigation path be preserved The history can be obtained directly from the NavPathBean forwarded by the DataLinkServlet param branch Branch to view param strNavVtxName Nav vertex of the selected branch param curVtxAttrsBean Attri
154. s activated by checking the baselines check box beneath the image windows the node viewer displays the 133 Figure 7 10 The node viewers baselines visualization baselines as blue cross hair targets labeled with the corresponding node s ID number as shown in Figure 7 10 7 1 4 The New Web Interface Has Support For Radiance En coded Images The pre existing web interface displayed jpg copies of the images in the City scanning dataset because most web browsers are not capable of decoding the rgb image format that is used to encode images in the City dataset This is no longer the case for the extended web interface which is now equipped with a Java class for decoding and displaying images in the rgb image format that can be plugged in to any of the visualization applets This class is also capable of re mapping the log radiance pixel values that are used in the City images to a linear scale The 0 255 scale of RGB color values in a standard image does not span a large enough range to fully capture all of the brightness values that are encoded in the high 134 dynamic range images in the City dataset Instead in the City images the 0 255 color component values of a pixel are actually the log of the pixel s radiance value for each of the RGB color channels The following formula is used to reconstruct a radiance value from its 0 255 log scale value R eRmint rg min 7 4 In this formula Rmin and Rmaz are the
155. s complexType gt lt xs element gt lt xs element name root gt lt xs complexType gt lt xs attribute name name type xs string use required gt lt xs complexType gt lt xs element gt lt xs sequence gt lt xs complexType gt lt xs schema gt 163 164 Appendix C An Implementation of Toms Method For Converting Between Geocentric and Geodetic Coordinates include ellipsoid hpp include coord_lib h I I II STITT constructor IIIIIIIIIII Construct a new Ellipsoid with given major axis and flattening rMajorAxis Earth s equitorial axis rFlattening The flattening is defined to be a b b where a is the semi major axis equatorial axis and b is the semi major axis Earth polar axis Ellipsoid Ellipsoid Real rMajorAxis Real rFlattening m_rMajorAxis rMajorAxis m_rFlattening rFlattening 165 m_rMinorAxis m_rMajorAxis 1 0 m_rFlattening m_rEccSq m_rFlattening 2 0 m_rFlattening m_rEccPrimeSq m_rMajorAxis m_rMajorAxis m_rMinorAxis m_rMinorAxis m_rMinorAxis m_rMinorAxis IIIIIII LLAtoXYZ I II III Transforms a geodedic lat lon alt coordinate in radians to a cartesian coordinate Both coordinates must be expressed relative to this ellipsoid Vector3D Ellipsoid LLAtoXYZ const LLA amp lla const Real rLon lla lon lt PI lla lon lla lon 2 PI Real rSinLat sin lla lat
156. s this by building a friendly interface to the Proj 4 cartographic projections libraray Though somewhat denigrated for its clumsy interface earlier in this chapter Proj 4 s virtue is the over 100 cartographic projections it supports There are too many of these to list here but the Proj 4 documentation provides a complete description of them Warmerdam 2000 Carrying out cartographic projections with Proj 4 the same type of interface that should be familiar with by now Each cartographic projection class in the Geospa tial Coordinate Transformation Library descends from a common base class called MapProjector Each of the different map projector implementations may require slightly different initialization parameters if any which are usually some type of in dication of the geographical region of interest or the units in which to output the results but once created they all share the same method for performing projections Vector3D project const LLA amp lla const The output of any cartographic projection is two dimensional so the z value of the vector returned will always be zero as I mentioned earlier the reason that the output is placed in a 3D Vector is to facilitate the use of the projection data in 3D applications Also as stated in Section 6 4 1 the sub classes of MapProjector 119 that implement individual transforms have their own typedefed monikers for the Vector3D output of the project function The MapProject
157. s time to being learning how to write a data model 3 3 1 Data Vertex Type Definitions Before describing how the the various elements of the data hierarchy are related it is first necessary to define all of the different types of data that might be found in the dataset As mentioned earlier this is accomplished with data vertex type definitions which have the form lt vertexType name data vertex type name gt lt Attribute definitions go here gt lt attributes gt lt integer attribute definitions gt lt string attribute definitions gt lt state attribute definitions gt lt attributes gt lt data signatures go in this section gt lt vertexType gt As this outline shows a data vertex may contain attribute definitions of three types integer attributes string attributes and state attributes and one or more stg nature definitions which tell the DataLink Server how to recognize a datum of this data type A data vertex type element can have zero or more attribute definitions of each of the different attribute types but the attributes must appear in the same order they are listed in the vertexType outline above That is to say all the integer attribute definitions must appear first the string attributes must appear next and the state attribute definitions must be last This is an artifact of the XML schema language that is used to define the model document type which has no construct for
158. s to a point x y in the view from the source camera If the transformation between the orientations of the two cameras is given by the matrix M then the equation of the epipolar line in the second camera s reference frame is r MP 7 2 M is a homogeneous matrix with a rotational and a translational component so the transformed line will have the form a t b1 agt ba agt b3 which after applying the perspective transform looks like 2 et 7 3 azt b ast bs 2In case the reader doesn t know or needs a reminder the parametric form of a line in three space is fr pi tp 129 The vanishing point occurs where t is infinite and is therefore al a3 a2 a3 7 1 3 New Global Visualizations The extended interface also has some new global visualizations which work in both the Map Viewer and the node viewers These allow the user to explore the vanishing points and baselines extracted during post processing The Vanishing Points Visualization If you ve ever driven on I 80 through Death Valley you are familiar with vanishing points In any perspective projection parallel lines in three space like the curbs of a highway converge to a point in the projected image called the vanishing point The rotational post processing stage aligns nodes in the City Scanning Project Dataset by attempting to align vanishing points that are visible from a stereo pair of nodes The vanishing points visualization is mea
159. s will be completly lowercase in order to differentiate between data and navigation vertices which have the same names The names of vertices in the actual model files do not have to adhere to this convention and they may contain any virtually any character apart from some special characters which are described in Chapter 3 36 more and more narrow When trying to map attributes to physical data in the data model however the system will never be concerned with the children of a vertex Instead it will need to examine the vertex s ancestors in order to determine the attributes it may have or it may have to enumerate the ancestors to reconstruct a path to the physical data Interactions Between the Data and Navigation Models The dashed blue lines in Figure 2 2 show one way in which the navigation model interfaces with the data model to retrieve the physical data The specification for navigation vertices which map directly to physical data in the data model include a type field which gives the name of the corresponding data vertex type in the data model When the navigation layer explores a child vertex of the currently selected vertex as shown by the green arrows in Figure 2 1 which is mapped to a data vertex type T to see which elements of type T if any should be made visible to the presentation layer it makes a request to the data layer for all branches of type T which are consistent with the current set of attribute values The data
160. saic etc A state attribute of the pose_refinement data vertex type which represents one of these pose directories is used to make the data s state of post processing refinement a navigable property lt vertexType name pose_refinement gt lt attributes gt lt stateVar name refinement gt lt state name initial gt lt signature gt initial lt signature gt lt state gt lt state name mosaiced gt lt signature gt mosaic lt signature gt lt state gt lt state name rotated gt lt signature gt auto_rot lt signature gt lt state gt lt state name translated gt lt signatures searched in the order listed here gt lt signature gt all_trans lt signature gt lt signature gt trans_set4 lt signature gt lt state gt lt stateVar gt lt attributes gt lt signature gt t refinement lt signature gt lt vertexType gt ol As you can see the refinement attribute has four abstract states initial mosaiced rotated and translated representing the four possible states of post processing refinement These abstract state names are not identical to the names of the pose refinement directories from which they get their values initial mosaic auto_rot all_trans and trans_set4 but the lt signature gt declarations map the abstract states to the appropriate directory names The precise definition of this mapping is is given by the pose_refinement vertex s signature lt signature gt t refinement lt
161. se tools allows the web author to create custom URLs for servlets and JSP deployed in a webapp as has been done here for the DataLinkServlet 74 Server treats any additional arguments to the request which do not correspond to attribute names as additional parameters for the content handler of the selected vertex and passes them on to the content handler without interference The author of a content handler should never have to write a request like this by hand The DataLinkDispatcher provides an API for formatting these request auto matically given an appropriate set of arguments and this chapter shall not make a mention of the request format again but before proceeding any further it is necessary for the reader to understand the operation of the history parameter because it is crucial in maintaining smooth navigation through the navigation hierarchies 4 2 1 The history Parameter The value of the history parameter is an ordered list of the navigation vertices that have been visited along the current navigation path The primary reason that the DataLink Server needs to maintain a history is to ensure that a user can never visit the same navigation vertex twice along a single path through the navigation hierarchy when the navigation model contains cycles Cycles are usually introduced into the navigation model by cross references between navigation vertices which are used to make it possible for the user to invert the order in which a pair
162. serverSocket private Thread m_listenerThread private boolean m_bShutdown private Vector m_vConnections private String m_strName constants timeout for socket waiting for connection private static final int SOCKET_TIMEOUT 1000 Helper class that encapsulates a Socket and its OutputStream private class Connection public Socket m_socket public OutputStream out public Connection Socket socket m_socket socket try out socket getOutputStream catch I0Exception e IIITTTTTTT TTT constructor I II III II III x Creates a new QutputStreamServer which listens for client connections on port number lt code gt nPort lt code gt 170 param nPort Port number to listen on public OutputStreamServer int nPort String strName m_nPort nPort m_strName strName I I II III 11 1 III III III III III VIII III II III III III III III TI methods 11111111L III I I start III x Starts the server public void start throws IOException m_bShutdown false m_vConnections new Vector m_serverSocket new ServerSocket m_nPort m_listenerThread new Thread this m_listenerThread start IIII stop I I II x Sends server the shutdown signal and returns public void stop m_bShutdown true I III I II disconnect I II III II Closes a connection does no
163. signature gt Make note that the vertex s signature is different thing than state attribute signatures In this example the vertex signature is a pattern specifier which says that a pose_refinement directory s name is equal to the value of the state attribute refinement Whatever state signature this value matches deter mines the abstract state of the refinement attribute The next section will explain this mechanism in detail but the important thing to glean from this discussion is that according to the above vertex definition when the refinement attribute has the value auto_rot the DataLink Server recognizes rotation as the abstract state of the refinement attribute and similarly for the other states and state signatures Notice that the translated state has two signatures all_trans and trans_set4 This is because the City scanning dataset contains data from multiple runs of the post processing algorithms Instead of overwriting the old data each time the algorithms are run the author of the new data may sometimes rename the old directories for safe keeping As a byproduct of this practice there are some nodes in the City dataset whose translated camera files are stored in all_trans directories some whose translated data are stored in trans_set4 directories and some nodes that have both all_trans and trans_set4 directories In general when the DataLink Server is searching for data with an abstract state that has more than one sign
164. sition estimates acquired using differential GPS The disadvantage of This type of coordinate system is also sometimes referred to as an East North Up ENU or East North Height ENH coordinate system Hofman Wellenhof et al 1997 20 this coordinate system however is that it is used only with the City group For this reason the position information in the City group s LTP coordinate system is not as useful to people browsing the data from outside the City group who work in other coordinate systems as it could be if it were exported in a more well known coordinate system like Earth Centerd Earth Fixed ECEF or lattitude longitude altitude LLA Hofman Wellenhof et al 1997 Similarly there is much data generated outside of the City group that would be of use to its members For example there exist detailed floor plans and maps of the MIT campus in AutoCAD format maintained by the Department of Facilities which could be used to check data derived by the City Group s post processing algorithms These maps and other outside data sources however are usually expressed in one of the commonly used cartographic coordinate systems such as US State Plane Registering the LIP coordinate system with the coordinate systems used in these maps would help the City group validate its data It could also be beneficial in merging the data we have collected about the building exteriors with detailed information about the building interiors Th
165. sses derived from the GeoDatum base class This class provides two method for converting between geodetic and geocen tric coordinates LLAtoXYZ for converting geodetic to geocentric coordinates and XYZtoLLA for converting geocentric to geodetic The prototypes for these methods are 1 Vector3D LLAtoXYZ const LLA amp LLA const 2 LLA XYZtoLLA const Vector3D amp v3 const The source code for GCTL s implementation of geocentric geodetic conversions is located in Appendix What follows is a terse mathematical summary of the steps involved It makes use of some additional ellipsoid constants which can be computed from the reference ellipsoid s semi major axis a semi minor axis b and geometric flattening Eq 6 2 e f 2 f 6 3 az _ b2 e 72 6 4 The procedure GCTL uses for computing geocentric coordinates from geodetic coordinates where latitude 0 longitude and h altitude is 10Rq 6 3 is the eccentricity of the ellipsoid squared 113 0 O lt T bU Change the range of longitude to be 7 to 7 0 27 0 gt r 2 I Ta Compute Earth s radius at the specified point e e Sin 3 2 R h cosdcos Compute x 4 y R h cosdsin Compute y 5 z R 1 e h sind Compute z Toms Method For Converting Geocentric to Geodetic Coordinates There is no closed form formula for converting from geocentric to geodetic coordinates all existing methods for performing this transformat
166. st way to demonstrate how to do this so let s begin by seeing how we would create a data model for the classical music dataset from Chapter 2 A diagram of this dataset is duplicated for convenience in Figure 3 1 First let us use the information from the previous section to create an explicit definition of the data itself 90 lt dataModel gt lt This URI specifies where the dataset lives gt lt home href usr local music_libs gt lt data vertex types gt lt dataset directory gt lt vertexType name dataset gt lt attributes gt lt string name datasetName gt lt attributes gt lt signature gt s datasetName lt signature gt lt vertexType gt lt composer directory gt lt vertexType name composer_dir gt lt attributes gt lt string name composerName gt lt attributes gt lt signature gt s composerName lt signature gt lt vertexType gt lt music directory gt lt vertexType name music_dir gt lt signature gt audio_files lt signature gt lt vertexType gt lt biography directory gt lt vertexType name bio_dir gt lt signature gt bio lt signature gt lt vertexType gt lt audio file gt lt vertexType name mp3 gt lt attributes gt lt int name opusNum gt lt string name pieceName gt lt attributes gt lt signature gt s s op 4d mp3 composerName pieceName opusNum lt s ignature
167. t remove it from connection vector 171 x param connection Connection to close private void disconnect Connection connection try connection out write CLOSING CONNECTION n getBytes connection out flush catch Exception e try connection m_socket close catch Exception e II III alive IIII N Checks whether the server is running by checking the current system time against the thread s timestamp If the server hasn t timestamped for a period longer than two socket timeouts lt code gt alive lt code gt returns false Otherwise it returns true return true if the server has updated its timestamp within the duration of two socket timeouts from the present system time x x x He KF KF KF OF public boolean alive if System currentTimeMillis m_lLastWakeupTime gt 2 SOCKET_TIMEOUT return false return true IIIIIII getPort IIIIII Returns the port number that this server listens on x return the port number that this server listens on 172 public int getPort return m_nPort EL OutputStream methods LLLLLLTTTTATT TT TT TTA TT TTT TTT TTT TTT TT IIIS write II II I public void write int b throws IOException Connection connection synchronized m_vConnections if m_vConnections null return for int n 0 n lt m_vConnections size n connection Connection m_
168. te directions In Fig ure 2 2 the edge going from the PIECES data vertex to the MUSIC_FILES vertex in dicates that MUSIC_FILES is the parent directory of the audio files which correspond to MUSIC_FILES This means that pieces inherit all the attributes of MUSIC_FILES COMPOSER and COMPOSER_DATASET the ancestor vertices of PIECES so a piece in the classical music library has values for the datasetName composerName opusNum and pieceName attributes In Figure 2 2 the attributes for a given data vertex type are listed under each instance of that vertex type The edge from the opus number navigation vertex to the pieces vertex in the navigation model on the other hand means that pieces is the child of opus number any audio files visible to the user located at vertex opus number in the navigation hierarchy must have the same opus number as the branch of opus number that the user selected to view The reason for this disparity is that it makes the implementation of the models in the DataLink Server much simpler and more efficient Chapter 5 which discusses the implementation of the DataLink Server will elucidate the logic behind this design but for now the basic intuition is that the user will navigate a hierarchy from the top down accumulating attributes which make the subset of the data in consideration Throughout this thesis I have adopted the convention that data vertex names will be written in UPPERCASE whereas navigation vertex name
169. the City LTP system One advantage of using a vector format map is that it makes it possible to zoom in without aliasing problems Using the new map viewer the user can now specify a viewing region by dragging out a rectangle on the screen with the mouse This allows the viewer to get a close up view of buildings sidewalks and other structures to see where the nodes lie in relation to objects in the physical world The user can then examine the accompanying view from the node s eye in the Mosaic Viewer to see how items on the map look from the node s point of view Right clicking zooms back out again and if the legend ever gets in the way it can be dragged and dropped to a new location Figure 7 2 shows the Map Viewer zoomed in to get a closer view of the nodes in the courtyard around the Green Building building 54 Notice that the Map Viewer is pretty smart about zooming No matter how elongated a rectangle the user draws it maintains the proper aspect ratio in the zoomed in view the zoom is scaled to fit the largest dimension of the rectangle that the user draws Also the Map Viewer automatically numbers nodes when the view is zoomed in enough that labeling the 122 Show baselines FT vanishing points f adjacencies g 6 8640 449 1720 Figure 7 1 The new Map Viewer 123 Figure 7 2 Screen shot of the Map Viewer zoomed in on the Green Building nodes nodes won t clutter up the image 7 1 2 Extensions to the
170. the case of the DataModel hash table lookup is also used to locate DataVertexType objects which hold the signatures and attribute definitions of individual data types Hash functions are exploited throughout the DataLink Server s implementation to speed up algorithms This is crucial in many places because the number of branches and attributes under consideration balloons rapidly The following sections will illu cidate a number of these tricks 5 2 DataLink Server Requests Come in Two Fla vors Section 4 2 gave a glimpse of the format of DataLink Server requests This section will explain them in detail There are two types of DataLink Server requests view IOne can do a little better bidirectional search which has the time requirement O 2b2 2 O b4 at an increased cost in spacial requirements Korf 1988 88 requests and view resource requests 5 2 1 View Requests View requests ask the DataLink Server to retrieve the branches for all of the children of a navigation vertex if any that are consistent with a particular set of attribute bindings and display the content handler of the selected vertex The content handler for a vertex which corresponds to a resource is the resource itself View requests have the following format http city lcs mit edu 8080 data2 DataLink view node amp dataset all nodes fixed amp nodeld 75 amp history dataset all_nodes In this example the URL to the DataLinkServlet is g
171. the graphs is something that the the models have to do several times for every request that is serviced by the DataLinkServlet and as the number of levels in either the data or 87 navigation hierarchies increases so do the number of lookups grow in an exponential fashion This means that brute force search through the models is not an option because these individual searches themselves would be inherently slow searching through a network really amounts to searching through a tree and the fastest time anyone can hope to do that in is O b7 where b is the branching factor of the graph and d is the depth in this case the number of levels in the hierarchy Korf 1988 Fortunately the fact that the DataLink Server searches for vertices which are guaranteed to have unique names means that it can to do a lot better if the graphs are each backed by a hash table which allows the DataLink Server to lookup a vertex by name in constant O 1 time Both the DataModel and the NavModel employ this strategy In each of these structures the vertices are represented by atomic vertex objects each of which maintains an array of pointers to its children The models can locate a vertex object in the hash table using the name of the vertex as a key Then to explore the vertex s children or ancestry in the case of the DataModel the model simply has to follow the pointers stored in each vertex to walk the edges of the graph like a linked list In
172. the specified servlet or JSP The developer can then interact with the JSP or servlet as they normally would via the web browser and any debug mes sages in the code will be redirected to a console on the developer s machine via the OutputStreamServer As mentioned earlier a single OutputStreamServer will accept multiple connec tions The MonitorServlet will only allow the developer to start one OutputStream Server on the remote machine at a time This is to prevent multiple unaccounted for OutputStreamServer instances from accumulating on the server and wasting re sources The OutputStreamServer periodically updates a shared variable which the MonitorServlet uses to check whether it is alive so it would be uncomplicated to ex tend the system to support multiple OutputStreamServers which could be remotely maintained by the MonitorServlet The source code for both of these tools is located in Appendix D The City server will only accept connections from within the Graphics Group domain 139 140 Chapter 8 Contributions and Concluding Remarks The principle contributions of this thesis are the DataLink Server a scaleable mod ular and customizable tool for building web interfaces to large datasets which auto matically provides navigation and data retrieval infrastructure and the Geospatial Coordinate Transformation Library which provides static C library routines for performing nearly any coordinate transformation invol
173. tic final void sortBranches Branch pBranches I III III III II TTT formatHistoryPath 1111111111111111 x x Formats a history array as a directory path string delimited by lt code gt lt code gt characters param pHistory History array to format x return The history formatted as a directory path string public static final String formatHistoryPath String pHistory 158 Appendix B A Complete Listing of the Model File Schema models xsd lt xml version 1 0 encoding UTF 8 7 gt lt xs schema xmlns xs http www w3 org 2001 XMLSchema elementFormDefault qualified attributeFormDefault unqualified gt lt root node gt lt xs element name modelDefs gt lt xs complexType gt lt xs sequence gt lt xs element name dataModel type dataModelType gt lt xs element name navModel type navModelType gt lt xs sequence gt lt xs complexType gt lt xs element gt lt data model type definition gt lt xs complexType name dataModelType gt lt xs sequence gt lt home specifies physical location of data Paths gt lt beginning with will be interpreted with respect to the gt lt webapp s dir gt lt xs element name home gt lt xs complexType gt lt xs attribute name href type xs anyURI use required gt lt xs complexType gt lt xs element gt 159 lt vertex type gt lt xs element name vertex
174. together to form a graph these attributes will be inherited Let s do that now lt dataModel gt lt data vertex declations gt lt vertex name DATASET type dataset gt lt vertex name COMPOSER type composer_dir gt lt parent name DATASET gt lt vertex gt ot lt vertex name MUSIC_FILES type music_dir gt lt parent name COMPOSER gt lt vertex gt lt vertex name BIO_DIR type bio_dir gt lt parent name COMPOSER gt lt vertex gt lt vertex name PIECES gt lt parent name MUSIC_FILES gt lt vertex gt lt vertex name BIOGRAPHY gt lt parent name BIO_DIR gt lt vertex gt lt vertex name PORTRAIT gt lt parent name BIO_DIR gt lt vertex gt lt dataModel gt See how the parent command has been used to specify the edges in the data model That s all there is to it The structure of the vertex element has left open the possibility of vertices having more than one parent This is not currently supported by the DataLink Server because allowing only one parent per vertex greatly simplifies the computational complexity of the algorithms used in the DataLink Server s imple mentation see Chapter 5 Room has been left in the DataLink Server code for the possibility of this extension however 3 4 Writing a Model To Direct Navigation Now that you know how describing the data is accomplished you can learn how to make it navigable This is the purpose of th
175. tputStreamServer which accepts telnet connections on the supplied port author ZMB public final class MonitorServlet extends HttpServlet private int m_nPort 9999 default port we will only support one server instance for now to prevent the creation of rogue servers private OutputStreamServer m_OSServer private Object m_serverMutex private PrintStream m_oldSysDutStream I II III init III public void init m_serverMutex new Object I II doGet I I II public void doGet HttpServletRequest request 176 HttpServletResponse response throws ServletException IOException doPost request response II III doPost IIITTTTT public void doPost HttpServletRequest request HttpServletResponse response throws ServletException IOException String strAction PrintWriter pw response getWriter response setContentType text html pw println lt HTML gt pw printin pw printin pw printin pw println pw printin lt HEAD gt lt TITLE gt Servlet and JSP Monitor lt TITLE gt lt HEAD gt lt BODY BGCOLOR FFFFFF gt lt Hi gt Servlet and JSP Monitor lt Hi gt action if strAction request getParameter action I null if strAction equals Go String strTestURL request getParameter testURL if strTestURL equals pw printin lt P gt You must supply
176. ude a complete listing of sibling branches for impli cated navigation vertex Consequently when the BranchHistoryScripter generates a context menu for a suppressed vertex it only provides the user with two options One of these takes the user back to the branch they selected The other is labeled choose another and returns the user to the selected branch of the parent of the suppressed vertex Thus if the user activated the context menu for the node vertex in Figure 4 1 for which enumeration is suppressed the only two options that it would present would be Node 67 the currently selected branch and choose another Selecting choose another would take the user back to the all nodes navigation vertex where the user would again be presented with a set of nodes to browse The Context Menus Use Javascript and DHTML Javascript is the medium of choice for creating the navigation context menus because it is low tech enough to be compatible with most browsers unlike Flash and Java and more importantly it is lightweight enough to avoid compromising the perfor mance of the presentation layer The BranchHistoryScripter provides two meth ods which are used to generate the Javascript which creates these menus at runtime in DHTML hence the name Scripter The first of these insertMenusScript Although I have used Java extensively throughout the City Data web portal I have steered away from the Java
177. ust the saturation threshold value The source code for the rgb radiance encoded image class and an rgb image 135 viewing applet can be downloaded from the City Scanning Project Dataset web site More information on the rgb image format can be found in Haeberli 2002 and the reader can learn more about radiance from Horn 1986 7 2 Content Handlers Can Perform Data Caching The DataLink Server makes it possible to cache some data that is accessed frequently but is slow to generate A couple of examples of this behavior exist on the browse page This page is the first navigation vertex that the user will traverse when acessing the dataset It brings up the Map Viewer of campus and plots the locations of all of the nodes in the dataset color coded according to state of post processing refinement The information about node locations comes from the camera file of image 0 for each of the nodes This means that without some kind of caching the DataLink Server has to retrieve each of the image 0 camera files individually open them and extract the pose data This is a very slow process because the number of these camera files is very large The bottleneck that comes from retrieving and opening a large number of files like this is not the DataLink Server The problem is actually the overhead that comes from making so many HTTP protocol request to open the files Caching can be used to cut down on this overhead significantly by packi
178. ute values given by lt code gt attrMap lt code gt which must lt b gt not lt b gt be null This method is intended for use in applications where Java 1 2 may not be used param strNavVtxName Name of the nav vertex to view Kr rt amp amp FF iF 150 param attrMap A set of name value pairings that define the attributes of the selected vertex param pHistory An array containing the names of the n vertices visited thus far which pHistory 0 pointing to the first vertex the root and pHistory n pointing to the latest x x return The URL for a propererly formated request to view nav vertex lt code gt strNavVtxName lt code gt constrained by the attribute values given by lt code gt attrMap lt code gt which must lt b gt not lt b gt be null x since JDK1 0 public static URL formatViewRequestURL String strNavVtxName SimpleNamedValueMap attrMap String pHistory SIITLITLTLTT TT TT TTT TTT TTT formatViewResourceRequest IIIIIIIIII III III III III Returns a String containing the URL for a properly formatted request to the DataLinkServlet for the resource given by lt code gt branch lt code gt at nav vertex lt code gt strNavVtxName lt code gt param branch The Branch which specifies the resource to request param strVtxName Nav vertex of the requested Branch can be obtained from the NavPathBean forwarded by the DataLinkSerlvet
179. ute cos of latitude 9 R oa Compute Earth s radius at the specified point 5 x 0 y4 0 z gt 0 10 d 7 z 0 y 0 z2 lt 0 Compute latitude tan mg 0 W gt x 0 y gt 0 11 0 3 silted Compute longitude 0 x 0 y 0 tan 2 o w aa cos gt L 12 h RaT cos lt L Compute altitude AT R e 1 0 W This makes use of two constants that Toms defines K 1 0026000 and L which is the cosine of 67 5 degrees The conditionals in steps 10 and 11 are to check for special cases when the point is at the equator or the poles In the actual implementation these are checked earlier in a trivial rejection step For a better understanding of why this method works please see Toms 1996 6 4 4 GCTL Can Do Datum Shifts The XYZtoLLA and LLAtoXYZ methods provide a way to switch back and forth be tween geodetic and geocentric coordinates in the same datum but sometimes it is also necessary to shift between two different datums especially when prepar ing to do a cartographic projection as some of these are only defined for certain datums Performing a datum shift on LLA coordinates directly is hard but it can be done using the Molodensky approximation or multiple regression equations EUROCONTROL and IfEN 1998 A simpler method can be derived however by making the observation that when coordinates in two different datums are expressed 115 geocentrically the only difference between the two datums is the position and orienta
180. vConnections get n try connection out write b catch Exception e disconnect connection m_vConnections remove n n end synchronized m_vConnections end write int b public void write byte b throws IOException Connection connection synchronized m_vConnections if m_vConnections null return for int n 0 n lt m_vConnections size n connection Connection m_vConnections get n try connection out write b catch Exception e disconnect connection m_vConnections remove n nas 173 end synchronized m_vConnections end write bytel b public void write byte b int off int len throws IOException Connection connection synchronized m_vConnections if m_vConnections null return for int n 0 n lt m_vConnections size n connection Connection m_vConnections get n try connection out write b off len catch Exception e disconnect connection m_vConnections remove n n end synchronized m_vConnections end write bytel b int off int len io Runnable interface 1111111111111111 Main thread loop Listens for incomming Connections public void run Socket socket Connection connection need to set the socket timeout or accept blocks until connection is made try m_serverSocket setSoTimeout SOCKET_TIMEOUT catch SocketException e oh well
181. ving points on the Earth or geographic maps These utilities have been used to build a powerful web interface to the City Scanning Project Dataset the largest collection of geospatial imagery in existence which exposes the data to anyone on the web including our colleagues in the field of machine vision 3D and 2D visualizations including the Map Viewer the Mosaic Viewer and the Epipolar viewer are used throughout the web interface to present the data in both a pedagocial and utilitarian way Briefly some high level contributions related to the DataLink Server are e The DataLink Server provides an easily configurable abstraction between a physical data hierarchy such as a directory structure or a database and a set of virtual navigation hierarchies e Content handlers create a way to embellish the web interface produced by the DataLink Server with a modular and customizable front end 141 e The XML modeling language for configuring the DataLink Server is a descrip tive and comprehensible language for describing data hierarchies and directing navigation through large datasets A short list of the services purveyed by the Geospatial Coordinate Transformation Library is e A universal tool for writing programs involving precise coordinate transforma tions of geographic coordinates including geocentric geodetic and cartographic conversions e An extensible API that makes coordinate types and reference frames explicit
182. with the set of attribute values determined by the current path through the navigation hierarchy When the user first connects to the system at the classical lib vertex no attribute values have been accumulated yet so any set of attribute bindings that specifies a composer_dir will be returned by the data layer The datasetName attribute has a value it s just been hidden from the user Now suppose that the directory named classical which houses the classical music dataset shares a parent directory with another dataset punk_rock punk_rock is the home of another music library with roughly the same structure as the classical dataset Inside punk_rock are some directories named after punk rock bands which are equivalent to composer directories in the classical dataset Now when the user visits the classical lib vertex the DataLink Server will output a listing like the one shown in Figure 3 4 Dead Kennedys et al are not the names of classical music composers In the list ing of Figure 3 4 the DataLink Server has branches for the composer directories from 61 Current Path classical lib Figure 3 4 Another listing of the classical lib vertex both datasets The classical branches represent sets of attribute bindings that look like datasetName classical composerName Johannes Brahms and the punk branches have attribute binding sets such as datasetName punk_rock composer The Clash Usually this t
183. x ample if the user clicked on the link for Node 2 node ID would bound to the value 2 This value persists as the user descends further into the navigation hierarchy so if the child vertices of V gt represent directories for pose data and images then they would only contain the camera files and images that belong to Node 2 in City dataset In order to resolve the mappings between the child vertices of V2 and the physical data in the dataset given the set of attribute values assigned by the user s current path through the navigation hierarchy the navigation layer consults the data layer as is depicted by the red arrows of Figure 2 1 The data layer examines the physical data and returns a set of branches handles to data which map to a given navigation vertex that are consistent with the current set of attributes to the navigation layer which in turn forwards these results to the content handler of Vj The content handler of V may then display them as links If the user clicks on one of these links the process is repeated anew for the newly selected vertex Path 2 depicts what happens when the user clicks on a link corresponding to a leaf vertex The mechanism for resolving this data is essentially the same as that described for Path 1 except that since a leaf vertex corresponds to a physical resource such as a file the data is simply served directly to the user instead of being passed through a content handler in the presentation lay
184. xs string use required gt lt xs attribute name value type xs string use required gt lt xs complexType gt lt xs element gt lt xs element name derivedProperty minOccurs 0 maxOccurs 1 gt lt xs complexType gt lt xs attribute name name type xs string use required gt lt xs attribute name source type xs string use required gt lt xs complexType gt lt xs element gt lt attribute redirection gt lt xs element name redirect minOccurs 0 maxOccurs unbounded gt lt xs complexType gt lt xs attribute name source type xs string use required gt lt xs attribute name destination type xs string use required gt lt xs complexType gt lt xs element gt lt nav vertex paramters gt lt xs element name param minOccurs 0 maxOccurs unbounded gt lt xs complexType gt lt xs attribute name name type xs string use required gt lt xs attribute name value type xs string use required gt lt xs complexType gt lt xs element gt lt xs element name branchName type xs string minOccurs 0 gt lt xs element name branchClassName type xs string minOccurs 0 gt lt xs element name contentHandler type xs string minOccurs 0 gt lt xs sequence gt lt xs attribute name name type xs string use required gt lt no type means nav specific vertex gt 162 lt xs attribute name type type xs string gt lt x
185. ype of behavior is desirable For example if the name of the composer lib vertex was changed to music lib the output of Figure 3 4 would be appropriate Suppose however that in this case we only want to make the classi cal library navigable One way of doing this would be to change the signature of the dataset vertex type to lt signature gt classical lt signature gt There are some cases however where there value of an attribute should be fixed for certain parts of the navigation hierarchy only One example from the City dataset is a show all rotated nodes vertex which would be the start of a navigation hierarchy for brows ing only nodes that have completed the rotation post processing phase This new hierarchy would be in addition to the regular hierarchies for browsing the nodes in all post processing stages and the way to implement it is with properties Attribute Values Can Be Fixed Using Properties A property declaration has two arguments the name of the attribute to set and the value to assign it The following listing shows how to fix the value of the datasetName 8The music library is to be made available to the employees of the British Consulate who s administration does not share the same views of Queen Elizabeth as The Sex Pistols 62 attribute using a property in the definition of the classical lib vertex lt vertex name classical lib gt lt child name composer gt lt property name datasetName

A Visualization Toolkit for Large Geospatial Image Datasets Zachary

Contents

Download Pdf Manuals

Related Search

Related Contents