SCS Undergraduate Thesis Topics
|Jared Go||James Kuffner||Real-Time Texture-Space Radiosity|
This research presents a new texture-space algorithm for computing real-time radiosity in static scenes containing both dynamic, high-frequency local lights and low-frequency environment lighting. This method only requires a texture map parameterization of the target model, which already exists for most models for the purposes of tangent space bump mapping and other techniques.
The algorithm presented generates a solution to the scene's diffuse global illumination by iterative solution of the Neumann form of the light transport equation, which enables us to dynamically simulate photorealistic illumination effects such as interreflection and occlusion. The scene radiance and corresponding transfer coefficients are efficiently encoded in texture space, which allows us to leverage graphics hardware to perform rapid Monte-Carlo sampling of the hemisphere at each texel, yielding an approximate solution to the transfer problem.
A key focus of this research is on the design and implementation of efficient sampling techniques to speed convergence at the low sampling and iteration rates required for real-time applications. The algorithm currently makes use of low-discrepancy sampling techniques and spatio-temporal filtering to achieve efficient iterative transfer while maintaining a low level of noise. Real-time performance on modern graphics cards is demonstrated, and an analysis of the convergence of the solution under different sampling and filtering techniques is also provided.