VRML Terrain Datasets SRI International

This page contains links to a number of large, multi-resolution terrain databases that have been generated in VRML97 format. These have been created as part of the terrain visualization work being performed at the Artificial Intelligence Center at SRI International.

Overview   |   Anchor/LOD Terrains   |   QuadLOD Terrains   |   Inline/LOD Terrains   |   Terrains With Site Models   |   EAI Interaction   |   Composite Datasets



OVERVIEW

The datasets are generated from elevation data and ortho-rectified imagery that are recursively subsampled to form a hierarchical pyramid, where each level of the pyramid is half the resolution of its parent. These levels are then segmented into a mosaic of small rectangular tiles and output in VRML97 format. A number of nested LOD nodes are used to produce a level of detail hierarchy where terrain that is closer to you appears in the highest detail while distant terrain appears in lower resolution.

All of these datasets were produced using the tsmApi library from SRI International. This library is freely available for users to download and use to produce their own terrain datasets.

If you don't have a VRML Browser installed on your computer, then you should be able to find a suitable one for your platform from the VRML Repository Browser Page. We recommend the use of Cosmo Player 2.1 with Netscape Communicator 4.x. You can download a copy of Cosmo Player 2.1 for Windows from here.



ANCHOR/LOD TERRAINS

These datasets use our Anchor/LOD technique for managing deep level of detail hierarchies. For VRML performance reasons, we split deep LOD trees into a number of separate files containing only a small number of detail levels (e.g. 3 or 4 levels) . The user is able to click over any of the higher resolution tiles in a scene in order to zoom into that region. Using these techniques we are able to produce terrain datasets in the order of many gigabytes such that a standard VRML browser is able to view the data interactively.

Yosemite Valley, CA

A 21x27 km region inside Yosemite National Park, CA, at 1 m resolution imagery. The data include Yosemite Valley, featuring El Capitan and Half Dome. This dataset contains a total of 11 levels of detail, with a total dataset size of 1.9 GB. The imagery for this dataset is 21135 x 27090 pixels (JPEG compressed).

Denver, CO

A 233x206 km region encompassing part of Colorado state, with 30 m resolution imagery. The city of Denver is toward the lower right corner of image. This dataset contains a total of 9 levels of detail, with a total dataset size of 73 MB. Elevations are exaggerated by a factor of 5. The imagery for this dataset is 6854 x 7755 pixels (JPEG compressed).

Camp Pendleton, CA

A 59x70 km region around the Camp Pendleton site in Southern California with 30 m resolution imagery. This dataset contains a total of 8 levels of detail, with a total dataset size of 6 MB. Elevations are exaggerated by a factor of 5. The imagery for this dataset is 1992 x 2317 pixels (JPEG compressed).

Fort Irwin, CA.

A 11x7 km region around Fort Irwin, with 1 m resolution imagery. This dataset contains a total of 12 levels of detail, with a total dataset size of 104 MB. The imagery for this dataset is 11362 x 6930 pixels (JPEG compressed).

Fort Benning, GA

A 3x2 km region around Fort Benning, with 15 cm resolution imagery. This dataset contains a total of 11 levels of detail, with a total dataset size of 166 MB. The imagery for this dataset is 18932 x 15533 pixels (JPEG compressed, grayscale).

LOD test dataset

This dataset is based upon the Fort Irwin terrain above but instead of imagery-mapped terrain, each tile is just a single, flat, randomly colored polygon. This can be interesting to see how the hierarchical levels of detail switch with distance. Total dataset size = 58 MB.



QUADLOD TERRAINS

Recently we have produced a number of custom VRML nodes in Java (e.g. QuadLOD and GeoTile) which explicitly manage all of the level of detail loading/unloading. QuadLOD performs on-the-fly loading and unloading of tile data, while GeoTile manages multiple terrain imageries and cultural features. These have the advantage that the user does not need to explicity click over regions to receive further detail; however, Java support in the current generation of browsers is not entirely stable and so these datasets can occasionally crash your web browser.

Colorado QuadLOD/GeoTile

A 233x206 km region encompassing part of Colorado state, with 30 m resolution imagery. This is the same dataset as above, except that we make use of the new QuadLOD and GeoTile nodes in order to provide fully automatic, progressive LOD.



INLINE/LOD TERRAINS

These terrain models just use Inline nodes in order to link all of the resolution levels into a single tree. These will only work if the browser delays loading of Inline nodes until they are switching into view by the LOD node. Unfortunately, the VRML97 specification is ambiguous with regard to this behavior; hence why we resorted to the Anchor/LOD approach above. Browsers that this technique will work with include: Cosmo Player 1.x (not 2.x) and Blaxxun Contact 4.x.

Yosemite Valley, CA

A 21x27 km region inside Yosemite National Park, CA, at 1 m resolution imagery. The data include Yosemite Valley, featuring El Capitan and Half Dome. This dataset contains a total of 11 levels of detail, with a total dataset size of 1.9 GB. The imagery for this dataset is 21135 x 27090 pixels (JPEG compressed).

The Earth

This a dataset of the entire earth produced with 1 minute resolution imagery and 30 second resolution elevation. The elevations have been exaggerated by a factor of 10. This dataset contains a total of 11 levels of detail, with a total dataset size of 1.2 GB (JPEG compressed). The imagery for this dataset was provided by Mountain Top Computing.



TERRAINS WITH SITE MODELS

We have produced a number of terrain models that include georeferenced site models, or other cultural features. The following links provide examples of terrains with building models, and weather simulations accurate overlayed. All of these currently use the Anchor/LOD terrain representation.

McKenna MOUT site

This is a small region of the McKenna MOUT site in Fort Benning. The imagery is 15 cm resolution imagery, and this site is not tiled. The elevation mesh is a TIN (triangulated irregular network). It includes building models with texture maps that were automatically extracted from the aerial imagery using SRI's Cartographic Modeling Environment (CME) product.

Fort Benning, GA

A 3x2 km region around Fort Benning, with 15 cm resolution imagery. This dataset contains a total of 11 levels of detail, with a total dataset size of 166 MB. The imagery for this dataset is 18932 x 15533 pixels (JPEG compressed, grayscale).

This dataset contains overlayed VRML site models (buildings, roads, and functional areas). These models were extracted automatically from SRI's Cartographic Modeling Environment (CME) product, and directly incorporated into the VRML terrain.

Denver, CO

A 233x206 km region encompassing part of Colorado state, with 30 m resolution imagery. The city of Denver is toward the lower right corner of image. This dataset contains a total of 9 levels of detail, with a total dataset size of 73 MB. Elevations are exaggerated by a factor of 5. The imagery for this dataset is 6854 x 7755 pixels (JPEG compressed).

This model contains an overlay of a Clear Air Turblence (CAT) isosurface that was recorded over this portion of the Rockies. The CAT VRML model was provided by the National Center for Atmospheric Research (NCAR), Boulder.

Camp Pendleton, CA

A 59x70 km region around the Camp Pendleton site in Southern California with 30 m resolution imagery. This dataset contains a total of 8 levels of detail, with a total dataset size of 6 MB. Elevations are exaggerated by a factor of 5. The imagery for this dataset is 1992 x 2317 pixels (JPEG compressed).

This models contains an overlay of wind vectors at 400m resolution that were supplied by the Naval Oceanographic Office (NAVO).



INTERACTIVE MANIPULATION WITH THE EAI

The External Authoring Interface (EAI) allows programs external to the VRML browser to control the content in that scene. Here we show the use of a Java applet running in one frame of the web browser to control the VRML plugin in another frame. (N.B. this will only work under the Netscape web browser.)

Fort Benning, GA

A 3x2 km region around Fort Benning, with 15 cm resolution imagery. This dataset contains a total of 11 levels of detail, with a total dataset size of 166 MB. The imagery for this dataset is 18932 x 15533 pixels (JPEG compressed, grayscale).

The EAI Interface at the top of the web browser lets the user select between different terrain imageries, and also choose whether they want to display the building models or not.



COMPOSITE DATASETS

All of the above examples illustrate a single terrain dataset in isolation. It is our goal to be able to represent composite datasets, where we can hierarchically insert higher-resolution datasets into a global model. For example, we may have 100 km imagery for the entire planet, then insert 1 km resolution dataset for the conterminous United States, and then insert a 1 m resolution dataset for the Yosemite Valley. Another issue here is to move away from flat terrains and take the earth's curvature into account.

Composite Bay Area Dataset

This dataset contains a model of the San Francisco Bay Area using 25m resolution color imagery over a region of 156 x 220 km. Embedded within this dataset there is a 1m resolution inset of the Menlo Park/ Palo Alto area, covering 12 x 14 km. Each dataset contains 10 levels of detail, for a total composite dataset size of 193 MB.

Composite Earth Dataset

This dataset contains a model of the earth at 4 levels of detail, built using a low-res 1024x512 image. Embedded in the globe are two icons that link to the Colorado and Camp Pendleton datasets (see below for details). Page down through the various Viewpoints to see different locations around the earth. Each of the 2 datasets have 3-D atmospheric data overlayed: a Clear Air Turblence isosurface over Colorado and wind vectors over Camp Pendleton. This is a prototype example showing embedded datasets in a global model.


Home page: http://www.ai.sri.com/VRMLSets
Please send any comments/questions to: vrml@ai.sri.com
Last modified: Sun Feb 6 16:03:29 PST 2000