Evolution of 3D Content

The visual quality of 3D models constantly evolves for real-time visual systems, with the goal of improving the simulated image such that it appears as close as possible to an entity's real-world depiction over multiple spectrums. This broad overview describes the evolution of 3D content for purpose-built image generators, arcade devices, and PC-based systems, with examples from the military and game markets.

This image is a composite of multiple modes of the Bearcat tactical armored vehicle model, which is included as part of MetaVR's vehicle entity library.

Initial practical development of 3D commercial visualization was for arcade video games. Real-time computational capability was limited at that time, such that only the manipulation of untextured and unshaded polygons was possible. The model appeared solely as a wire-frame in the game, as shown in the following example. This wire-frame depiction of an M1 tank model uses 263 triangles and has no texture map:

The modern image generators of the late 1980s had increased graphics and geometric processing capability and were able to display more realistic models. The next example shows the same model with a 128-128 pixel texture. The texture is applied to the model repeatedly to create the appearance of a camouflage pattern. In late 1996, this same model fidelity could be depicted with PC-based graphics cards with the introduction of the 3Dfx Voodoo chipset.

This texture was created in a graphics program; it was not derived from a photograph of a real-world vehicle. Using a repeating surface pattern on a model requires less texture memory in the 3D graphics card than a texture that has a one-to-one mapping of the surface detail with a photo-realistic texture pattern. The vehicle tracks were modeled by another non-repeating 256 x 64 pixels texture pattern that maps the image pixels one-to-one to the model features for the road wheels. The front tread itself was modeled with a repeating texture. Repeating textures are used to capitalize on a model’s symmetry in that symmetric features can be modeled by a single texture used multiple times. This model is a hybrid example of repeating and non-repeating textures with a total effective texture allocation of approximately 109 KB.

The next example is a similar model comprised of repeating and non-repeating textures. In this case, the geometric complexity is higher than in the previous example, with 441 triangles and a total texture size of approximately 500 KB.

The following model example uses 445 triangles and a semi-non-repeating 256 x 256 pixel texture map that represents a texture pattern that maps the image pixels one-to-one to most of the model features. Some symmetric parts of the model use the same area of the texture map for multiple parts of the vehicle, as in both the left and right side road wheels, which are represented by single component of the image. The top of the hull is modeled by a unique texture map for the entire surface of the hull. Road wheels are modeled as a skirt of non-moving geometry.

The texture, shown below to the left, was developed in a graphics software package with the intent to replicate the appearance of the vehicle. Photographs of the actual vehicle were used as a guide for creating the texture map elements in lieu of photographic textures. The picture of the CCTT M1 shown below to the right also uses this process:

The commercial game market has developed 3D model content in parallel with the visual simulation community. Most current games still use textures developed in a graphics drawing package in an attempt to replicate the appearance of the vehicle; typically the textures of the models are not made from photographic elements. The geometry of these vehicles is of very good quality and the higher-end games fully model the road-wheels and animate the motion of the tracks such as in this next example of an M1 tank featured in the game Operation Flashpoint.

Smoke effects from the engine are modeled, thereby adding significantly to the fidelity of the vehicle. The texture budget for game vehicles is typically limited; thus game designers must develop their products for a wide range of 3D graphics cards where the least common denominator may be 32 MB of texture memory.

The benchmark in the simulation community for 3D models has traditionally been the set of CG2 models in the Facets library.

The Facets M1 model shown above consists of two 512 x 512 texture maps created with a graphics drawing program. The 2,051 polygons that comprise the geometry of the vehicle is on the order of a factor of two more of the analogous game model. The road wheels are modeled as a skirt, not individually. Currently, game models and visual simulation models have parity in visual quality. However, game models such as this one have features that add significantly to the model realism that do not typically appear in most real-time image generators.

The non-realtime graphics world is mostly populated with visually stunning 3D models with very high polygon counts. For example, here is an M1A1 Abrams from the De Espona 3D Models Encyclopedia that is comprised of 130,000 polygons:

Models with this high of a polygon count are typically not practical to render in real-time games and simulations at this time. Note that the textures are hand-drawn, not photo-realistic. This community of model developers tends to equate realism with higher polygon counts.

For comparison with the previous models, a RealDB model of an M1 is shown in the next two examples. The model in the first example below is comprised of 2,013 triangles and uses a primary fully photorealistic 1024 x 1024 pixel texture map and secondary detail textures for a total texture allocation of 3 MB. This version of the model uses a static texture and non-articulated components to model the road wheels and tracks. 3D models that use photo-realistic textures provide a greater degree of realism when compared with artificially created textures.

The next example shows the same RealDB model but with fully animated and individually geometrically detailed road wheels that increase the final geometric complexity to 3,073 triangles. The texture allocation for this model is approximately 4 MB. Entity models are approaching sufficient detail for them to be used to evaluate real-time target detection algorithms in distributed simulation exercises.

The visual and geometric fidelity of 3D models will to continue grow in part driven by the standards of the 3D game market and the need for ultra-realistic vehicles for sensor feedback and visual identification in dense urban environments. Diverse subvariants of a particular military vehicle type are typically found in a given country's military (such as, the T-62, T-72, and T-90). Each vehicle then has a unique set of capabilities such as armor protection, counter-measures, and camouflage schemes. High-fidelity models enable one to distinguish subtle differentiating features between the vehicles.

To achieve the aforementioned fidelity, MetaVR has developed a photographically specific 3D model standard, which was used to build this latest version of the M1A2 3D model shown below:

This model is comprised of 11,533 triangles and 9.26 MB of textures (excluding textures of multiple paint schemes).

These final images compare MetaVR's M1 model with a European camouflage version of an actual M1 tank from which the MetaVR photospecific model was derived.

In the side-by-side images directly above you can compare the real M1 vehicle and the version of the MetaVR M1 model shown previously. The photograph of the actual M1 tank is courtesy of Army Recognition.