Collision Detection of TriangleMeshes using GPU

Abstract: Collision detection in physics engines often use primitives such as spheres and boxes since collisions between these objects are straightforward to compute. More complicated objects can then be modeled using compounds of these simpler primitives. However, in the pursuit of making it easier to construct and simulate complicated objects, triangle meshes are a good alternative since it is usually the format used by modeling tools. This thesis demonstrates how triangle meshes can be used directly as collision objects within a physics engine. The collision detection is done using triangle mesh models with tests accelerated using a tree-based bounding volume hierarchy structure. OpenCL is a new open industry framework for writing programs on heterogeneous platforms, including highly parallel platforms such as Graphics Processing Units(GPUs). Through the use of OpenCL, parallelization of triangle mesh collision detection is implemented for the GPU, then evaluated and compared to the CPU implementation