Current techniques of radiosity-based rendering are inadequate for complex scenes with 100,000 polygons or more. Existing approaches either take too long, or produce poor images. By fully exploiting the special organization of some recent graphics hardware, fast computers, and some new algorithms, we can generate accurate images far more quickly than existing application programs.

Our approach pushes the frontiers of both speed and correctness. It combines the speed of progressive radiosity with the flexibility provided by adaptive subdivision. We have developed a new visibility algorithm specifically designed to take advantage of the fast graphics hardware available on the Silicon Graphics IRIS machines.

Another difficulty addressed is the problem of having to start from poor input models. Many of the existing architectural models of buildings have been developed with 2-dimensional drafting tools such as AutoCAD. These models have various problems including randomly oriented faces, intersecting or poorly shaped faces, and overlapping coplanar features. With a number of preprocessing tools, such input files are converted into a more consistent description that is directly suited for radiosity analysis.

These experiments and improvements have been developed in an environment that permits us to replace and update individual program modules so that we can compare and evaluate, for instance, different patch and element meshing algorithms, and study the effect of different algorithms on the quality of the final image.





Download Full History