Before I go and turn my pages of handwritten notes into code, can anyone comment on this and let me know if I’m on the right track?
I need to test a bounding box against the current view frustum. I don’t need to differentiate between boxes that are partially in the frustum and totally in the frustum, they can either be rejected completely or they need further processing. Speed is, of course, of the essence.
So… here’s what I’ve come up with. I know how to code all this, I just want to know if it makes sense to you experts.

I extract the equations for the 6 planes of the frustum from a concatenation of the current projection and modelview matrices. I only do this if/when the “camera” moves.

Given a bounding box with 8 vertices, I loop through all 6 planes doing the following:

Loop through the 8 vertices, calculating the distance of the vertex from the plane. A positive distance means it’s in front of the plane (or outside of the frustum) while a negative value means it’s behind the plane (possibly inside the frustum). If I find a vertex that is behind the plane, then I can refrain from testing the remaining vertices (if any) because at least some part of the bounding box is behind the plane.

If all 8 vertices are in front of any of the six planes, then I can refrain from testing against the remaining planes because it’s totally outside the frustum.

If, for each plane, at least one vertex is behind the plane, then some portion of the box is visible and it cannot be rejected.
Does this sound about right?
Testing a bounding sphere would be similar, faster in fact, since I only have to test a single point and compare the distance from the plane to the sphere’s radius.
Toom
[This message has been edited by Toom (edited 12012000).]