posted April 27, 2000 09:01 PM
Sounds like the same problem found in ray-casting engines and voxel terrain. Let me see if I can dig up a few references right quick...
Outcast's GDC presentation (some voxel stuff): http://www.appeal.be/products/page1/Outcast_GDC/outcast_gdc_1.htm
Random Stuff (ok, not so good, but www.google.com is your friend ) http://www.cs.sunysb.edu/~vislab/projects/volume/Papers/ http://www.gamasutra.com/features/20000215/lander_01.htm
I would guess that for a game-level engine, you could pick the height sample under the object in question, then create a plane using two nearby height samples. This reduces the physics to point-mass and inclined-plane equations from basic physics. ie: how the ball will bounce off a given point of terrain.
As for line of sight, I recall a calculation that is begun at the eye and cast outward along the terrain checking each height sample for intersection. However this model was optimized for edge-on viewing of the terrain, not general view metrics.