3D Representation

With a suitable infrastructure in place to index georeferenced data, we need to discuss the details of how these data can be represented and generated. Accordingly, this section will describe the creation and maintenance of georeferenced data for the Digital Earth. Particular emphasis is placed upon the adoption of open standards and the ease with which new data can be generated and embedded into the global infrastructure.

Data Representation

The Digital Earth will consist of huge volumes of distributed data. We require that the user be able to interact with these, experiencing minimal download times and high frame rates. This is made partially possible with the existence of advanced high-speed, low-latency NGI networks; however, this is only part of the solution. A single computer could not possibly handle all of the data in the Digital Earth if it were transmitted as a flat structure. For example, color imagery for the entire world at 1 m resolution would require over 1 petabyte (10^15 bytes) of memory. If we had elevation data for the world at 30 m resolution, this would produce a geometric model of over 500 billion (5 x 10^11) polygons. It therefore becomes essential for us to implement some form of multi-resolution representation for our data, so that only the minimum amount of data at the lowest appropriate resolution need be transmitted to the client. The following sections describe how we propose to represent such multi-resolution terrain, imagery, and other features in VRML.

Multi-resolution Data

Multi-resolution terrain is best represented using a hierarchical structure such as a tiled, pyramid scheme. This involves progressively downsampling an image or elevation bitmap to produce the multi-resolution pyramid. Each level of this pyramid is then segmented into a grid of equally sized rectangular tiles, for example, 128 x 128 pixels. A tile at one level of the pyramid will therefore map onto four tiles on the immediately higher-resolution level, that is, the tiles at the higher-resolution level cover half the geographical area of the former. Using such a representation, we can progressively transmit higher resolution data around the area of interest while other regions remain in low resolution (see Figure 2. The use of tiling also allows us to fetch and display only those sections of the Digital Earth that are visible from a certain vantage point.

Figure 2 (a) A tiled, pyramid representation for terrain, and (b) Illustrating the use of this structure to produce viewpoint-dependent, multi-resolution imagery.

One of the principal goals of this proposal is to allow multiple types of georeferenced data to be embedded into the global infrastructure. For example, we could have 100 km resolution data for the entire globe but recursively insert higher-resolution datasets for smaller regions of interest, for example, a 1 km resolution dataset for the conterminous United States and a 1 m resolution dataset for Yosemite Valley, CA. Then we could insert 3-D models for buildings in Yosemite Valley, some of which could contain hyperlinks to various multimedia presentations about the National Park. Our data representation must be able to support this level of flexibility.

VRML Representation

We propose to implement the above Digital Earth representation by introducing three primary types of VRML files: Tree files, Terrain Tile files, and Data files. These are the basic building blocks of our VRML terrain representation.
Tree files
These implement part of the multi-resolution structure for the entire globe. In effect, these files are the glue that holds the Terrain Tile and Data files into a global quad-tree structure. The Tree files enable us to distribute the entire hierarchy over multiple servers and to abstract the multi-resolution structure from the actual terrain and other data. The tree files contain descriptions and links to all available data for a single region at one level of detail.
Terrain Tile files
These contain the actual terrain data for a single tile of a pyramid. This includes the elevation geometry and links to the texture-map imagery for that specific tile of terrain. There will exist a separate pyramid of Tile files for each terrain dataset. Using the previous example, the 1 km U.S. dataset and the 1 m Yosemite dataset will each have its own pyramid of Terrain Tile files.
Data files
These are arbitrary georeferenced data embedded into the Digital Earth structure. For example, these could be building models, roads, weather simulations, and terrain annotations. Any of these might also contain hyperlinks to other types of information such as audio, video, or textual presentations. This is where the richness of the Digital Earth is enabled.

Data Production Tools

For the Digital Earth to be a useful and widely adopted facility, it must be extremely simple for users to introduce their own data into the global infrastructure. This requires the existence of several tools to help manage the contents of the Digital Earth. We propose to develop software that will aid users to produce multi-resolution terrain in VRML format and insert these data into Digital Earth, to embed new georeferenced objects into the structure, to update existing objects, and other useful operations. We propose to utilize SRI's tsmApi library to fulfill these capabilities.

The tsmApi Library

The tsmApi (Tile Set Manager Application Programming Interface) is an open, freely available, C library provided by SRI International. The library offers functions for reading, writing, and managing terrain-related data. This includes functions for generating multi-resolution, tiled, terrain using VRML. Currently, these datasets can be created from supported input formats such as raw imagery, PBM images, and LAS bitmaps.

The tsmApi library also includes a fully re-entrant VRML parser written at SRI International. This can be used to parse VRML files efficiently into memory and to write these structures back out to a VRML file. Functions are also provided to perform various geographic coordinate transformations, for example, UTM to Lat/Long and Lat/Long to Geocentric. (These are based upon code from the National Imagery and Mapping Agency's NIMAMUSE product.)

Pre-compiled tsmApi distributions are available for several platforms as well as the full C source code and extensive Web-based documentation. Using all of these materials, users can develop their own tools for managing the content of the Digital Earth. We propose to deliver several such tools as part of this project, for example, software to perform the following operations:

  1. Produce multi-resolution tiled terrain pyramids in VRML.
  2. Insert these terrain pyramids into the main infrastructure.
  3. Embed georeferenced objects into the main infrastructure.
  4. Remove selected components from the Digital Earth.
  5. Modify existing components, for example, update metadata or change URLs.
  6. Query the contents of the Digital Earth.

Previous: Infrastructure
Next: Vizualization System

SRI's Digital Earth Project
SRI's Digital Earth Proposal

digital-earth@ai.sri.com -- Fri Apr 23 17:52:37 1999