Results
Several tests:
- intersecting or not,
- simple (4 tri.) or complex (1224 tri.) organ
- rigid or deformable organ
- moving or still tool
Several types of graphics hardware
Notes:
To validate this approach, we performed several tests, and we compare the time needed to detect collision with the RAPID public domain package which implement the OBB tree method.
Tests were performed in configurations where collisions occurred or not. We use several geometries of various complexity. From a single tetrahedron to our non convex liver mesh.
Collision detection was tested on rigid or on deformable organs. With our technique, there is no difference between the two. But for RAPID, it makes a difference. For the rigid object tests, we don’t take into account the RAPID pre-computation times. As we don’t have an easy way to update this data structure when the object deforms, we run the pre-computation at each time step for the deformable object tests.
These tests were performed both for a moving and a still tool.
We also use several processors and several types of graphics hardware.
We uses high performance OpenGL hardware, low cost gaming graphics board (no OpenGL), and we even run theses tests on computer with no graphics hardware at all, performing all the OpenGL calls in software thanks to MesaGL.