Graphics Reference
In-Depth Information
on their bounding boxes first, before filtering on the objects themselves. Because
the test does not rely on closed geometry, when a set of objects has been identified
as potentially colliding the faces of the objects can be treated as individual objects.
Reapplying the algorithm on the face sets returns the set of potentially colliding faces
upon exit. Knowing these faces, exact intersection points can then be computed on
the CPU if needed.
10.5
Summary
This chapter looked at utilizing the increasing processing power of GPUs for image-
space-based intersection techniques. When the GPU is used for such purposes, it
is important to restrict the communication between GPU and CPU because reading
back data to the CPU is often slow. The main solution to overcoming this commu-
nication bottleneck relies on using hardware-supported occlusion queries to pass
information back to the CPU. The tests presented in the chapter all use these occlu-
sion queries. An alternative solution is to perform buffer readbacks of buffers that
have been reduced in size through (possibly repeated) downsampling using bilinear
filtering.
Image-space-based overlap tests for both convex and concave objects were
presented, and inherent accuracy problems with image-space-based tests were
noted. Last, a filtering method to cull a set of objects down to a small set of
potentially colliding objects was discussed. This filtering method can be applied
both as a broad-phase filtering method to the objects' bounding volumes or as a
narrow-phase filtering method to the objects themselves.
