Our concluding example demonstrates the simulation of a complex plant scene. Here we applied some of the previously discussed instancing methods, in par-
ticular the approximate instancing procedure, whereby the terrain from Fig. 7.7 serves as a basis.
The parameterized data base of different species is generated using the Xfrog modeling system (see Fig. 8.15). In the scene shown below, the production of the apple tree model took several hours. All other objects were created more quickly, partially they could be extracted from other scenes.
|
|
|
For the rendering of the plant population, density functions were specified graphically. The stinging-nettle population in Fig. 8.16 serves here as an example. To parameterize the growth rate in this illustration, the height image was used, so that the specimens located in the lower river valley reflect a higher growth rate than those that are situated in comparatively dry regions on the small hills. Large objects, such as the apple trees, were added manually into the scene.
The already mentioned approximate instancing was applied to the completed populations. For each species a different number of representatives was selected. The values are indicated in the table of Fig. 8.17. The number of polygons without instancing amounts to 16.6 million for the scene; with instancing there are only one million. Thus, all in all, a reduction factor of 16.6:1 is obtained.
If further reducing the representatives, the reduction factor can be improved. In isolated cases, one representative is sufficient, so that the expenditure for the production of a scene is confined to the modeling and simulation of only one plant.
All in all, instancing procedures provide a good basis for the handling of geometrical data in natural scenes. The remaining data typically is of a size that is also manageable with standard computers.
Figure 8.17 Number of representatives and instances in the example scene
In the next chapter, we will introduce procedures for rendering the geometric data. in this context there are also approaches to additionally reducing the data with so-called level-of-detail modeling methods. While these approaches do not facilitate data reductions for the geometrical models of the representatives, they do, however, significantly reduce the data to be generated per image. We will demonstrate that this kind of methodology allows us the rendering of complex ecosystems in near real time.
|