| OCR Text |
Show CHAPTER 6 RAY INTERSECTION METHODS SOME EMPIRICAL RESULTS A single ray tracing program was modified to work with different ray intersection methods [27]. The initial approach used hierarchical bounding volumes similar to the approach in [30]. The next method was based on octrees [16], which was the subject of Chapter 5. The third implementation was based on Kay and Kajiya's method [24]. This allowed a comparison of the speeds of the three methods where some of the variations present in previously published comparisons [3] were removed. For the tests reported here, all of the methods were implemented on the same software testbed, and were run on a single computer with only one user. Under the original ray intersection method, objects were grouped hierarchically in an attempt to partition the set of objects to be tested against a ray. The implementation was similar to Rubin and Whitted's [30]. One implementation difference was that in this case the bounding volumes were three-dimensional bounding boxes aligned with the coordinate axes, instead of the parallelipipeds suggested in [30] . Manual clustering of surfaces into hierarchies was necessary in this method. However, B-spline surfaces were automatically converted into a binary tree hierarchy of triangles as they were subdivided. The ability to place surfaces into hierarchical groups was necessary to attain good performance. When a ray intersection was required, a ray was tested against the elements of a set of top level objects in the hierarchy. H there was an intersection of the ray with a top level bounding volume, |
