My Collision Detection system can refresh a hierachical tree consisting of 6k nodes representing 3k polygons, in 11.14 micro-seconds. Is that good? Do games usally have more polygons per model? and if the polygons count is super high would it be feasible to have special collision detection models? Thanks in advance

It works by making a 14-dop bounding box hierarchy tree structure. Where each bouding box has two sub bounding box, and so on and so on. Also I found a higher polygon model, so now 6k triangles, and it refreshes 17 micro-seconds. now that refresh also can handle if the model was deformed, rotated, translated, and scaled. Now I have a "quick refresh" that only takes 7 micro-secs for a 6k polygon model that can only handle translation. Now for that 6k model it creates 10k bouding boxes for intersections. Now you said games don't check triangle-triangle intersection?

Well it all depends on how deep the tree should be constructed. My professor who is using this for robotics, so he wants exact coordinates of collisions and ray collisions. Now for a video game maybe the develpers want to go down to a bounding box for every 2 triangles, or 3. Then of course there wouldn't be so many. But yes in a hierarchy tree the entire model does have its own bounding box, then the next level that box is split into 2, and then those boxes are split, and so on and so on. The tree is a "full" tree but it may be unbalanced. Meaning Every node has 2 children creating what is called a binary search tree, which makes intersection tests much faster. Also lucky for me I created a "system" class which contains certain number of theese hierachies, the developer can have multiple of these systems maybe organized by scene location to further speed up the testing process. I recently included a ray intersection as well. w00t my system is almost complete.

Promit has a good point. 6k node for 3k objects would be okay, but 3k polygons is going to mean that you have quite a small poly-budget for your art assets.

A given object (a player, say, using a 500 polygon model) should have one node in your collision detection tree; you don't need to reduce the collision detection to a per-poly level because

a) you (probably) don't need to test the player for collision against himself, and
b) you (probably) don't need polygon-accurate collision of other objects (bullets) with the player. Bounding volumes should suffice.

EDIT: The OP got a post in after Promit's while I was writing mine.

Quote:
Now for a video game maybe the develpers want to go down to a bounding box for every 2 triangles, or 3.

Even that is likely to be FAR to fine-grained of a resolution.

However, your statement about doing this for some robotics thing and needing exactly collision location absolves you from my statements about not needing polygon accurate collision, it sounds.

People plz read my replies. My professor wants a triangle-triangle intersection. THe tree could stop building if

a) the number of nodes suffice for accurate collision detection, i.e. if you
want 5 nodes per model, then hierarchy can do that.
b) The developer maybe wants to have the nodes do not go further down than 4
triangles per volume. Again this can be setup.

Lol the timing on forums is tricky somtimes. Hey if the less nodes then refreshing a tree could be reduce dramatically. I'm not even worried about intersection test cause a 14-dop tests are fast. THe tests are maybe slower than AABB trees becuase of the more planes to test, but beacuse k-dops represent volumes more accuratly tests that would normally traverse any other tree could be stopped early when seeing that the objects do not intersect.