BSP-trees (binary space partitioning trees), often in conjunction with a PVS (potentially (or possibly?) visible set).
Octree''s (or their lesser variant, quadtree''s) - a spatial space subdivision method. Construct the octree, recursivly test each bounding volume against the frustum (or a rough approximate of it, ie. a sphere bounding the view-frustum). If a bounding volume (node) is visible, all of the contained objects are, no further tested is required. All the child nodes are visible and thus also it''s contained objects. If the node is partially visible, perform further testing on the node''s children nodes. Start over at the previous step! If the node is not visible, it''s nor it''s children nodes objects are visible and thus no further testing is required - all geometry that resides inside the bounding volume can be discarded.

If that''s what you mean by geometry culling...?

If you, however, mean occlusion culling, I''m not very experienced... check out beamtrees, might be somewhere to start. Or a z-pyramid/octree approach, as discussed in Real-time rendering by Möller - a good source for further information on occlusion culling by the way.

Hope at least something helped...!

Cheers!

If you''re after occlusion culling, you should definitely check out hierarchical occlusion maps (HOMs). This is one of the most efficient realtime occlusion culling methods you can find.

- AH