Geometree

In November 2004, GeomeTree, a basic research project with support from the Austrian Science Fund (FWF) has been started. The goal of this project is to address the issues of geometric detail and rendering speed by building a multi-resolution representation of trees, that allows a seamless change of scale, all the way from microscopic levels (e.g. the user inspects the bark of a tree) to macroscopic levels (the user sees woods in the distance). As it is impossible to hold all the geometry for the entire scene in memory, it is imperative to develop new methods for generating the necessary data on the fly. Additionally, new methods for rendering will be devised to handle the large amounts of geometry that will be generated. The need for such new algorithms arises from the fact that convincing realism in real-time rendering of trees has not been achieved with current methods. 

In order to provide the necessary geometry it will be procedurally generated on the fly, at an adequate level of detail for the current viewpoint. In this part of the project we will investigate methods for procedural generation of plants with special emphasis on the multi-resolution nature of our desired output. The new methods will provide the following advantages: geometry on demand, a true multi-resolution representation, and volume-based visibility computation for occlusion culling.

Since the enormous amounts of geometry required for realistic plants cannot be completely rendered in every frame, new methods for real-time rendering of plants will be developed. Again the multi-resolution nature of the problem will be taken into account: as the viewer approaches small parts of a plant, additional geometry will be rendered in a seamless way to create the illusion of a coherent, detailed model of the plant. Here level-of-detail techniques and image-based rendering will be combined to a comprehensive new method for rendering multi-resolution models of plants: in the near field, polygonal meshes, generated on the fly will be used, in the far-field impostors will be used to amortize the rendering cost over multiple frames.

The project will result in a system for rendering trees in real-time, and in such a way that they can be viewed at a distance or inspected up close without loss of geometric detail.