Cornell University Program of Computer Graphics
Global illumination via density estimation.Peter Shirley, Bretton Wade, Philip Hubbard, David Zareski, Bruce Walter, and Donald P. Greenberg.
In Eurographics Rendering Workshop 1995. Eurographics, June 1995.
This paper presents a new global-illumination algorithm for highly complex static environments containing both diffuse and non-diffuse surfaces. The algorithm accounts for all types of surface reflection, accommodates textured surfaces, and supports coarse-grained parallelism. A key to this method is a novel decomposition of the problem into a sequence of three loosely-coupled phases. The first phase consists of Monte Carlo particle tracing in which power-carrying particles are emitted from each luminaire, and tracked through the environment until they are absorbed. A list of particle ``hit points'' is kept for each surface. In the second phase, called the ``density-estimation'' phase, the stored hit points are used to construct an approximate irradiance function for each surface. In the final phase, called the ``meshing'' phase, each surface irradiance function is approximated by a piecewise-linear function. Display output can be either Gouraud-shaded polygonal elements for interactive walk-throughs, or ray-traced pixels for higher-quality still frames. The method is modular, relatively easy to implement, has low memory overhead, and has produced view-independent display meshes for models larger than the ones that have been processed by previous methods. Such solutions could improve the state-of-the-art in architectural simulations, immersive environments, and other virtual reality applications.
This paper is available as a compressed Postscript file SWH+95a.ps.gz (304K).