Efficient Object-Based Hierarchical Radiosity Methods

Abstract

The efficient generation of photorealistic images is one of the main subjects in the field of computer graphics. In contrast to simple image generation which is directly supported by standard 3D graphics hardware, photorealistic image synthesis strongly adheres to the physics describing the flow of light in a given environment. By simulating the energy flow in a 3D scene global effects like shadows and inter-reflections can be rendered accurately. The hierarchical radiosity method is one way of computing the global illumination in a scene. Due to its limitation to purely diffuse surfaces solutions computed by this method are view independent and can be examined in real-time walkthroughs. Additionally, the physically based algorithm makes it well suited for lighting design and architectural visualization. The focus of this thesis is the application of object-oriented methods to the radiosity problem. By consequently keeping and using object information throughout all stages of the algorithms several contributions to the field of radiosity rendering could be made. By introducing a new meshing scheme, it is shown how curved objects can be treated efficiently by hierarchical radiosity algorithms. Using the same paradigm the radiosity computation can be distributed in a network of computers. A parallel implementation is presented that minimizes communication costs while obtaining an efficient speedup. Radiosity solutions for very large scenes became possible by the use of clustering algorithms. Groups of objects are combined to clusters to simulate the energy exchange on a higher abstraction level. It is shown how the clustering technique can be improved without loss in image quality by applying the same data-structure for both, the visibility computations and the efficient radiosity simulation.