Large Scaled Deformation Collision Detection Based On GPU

How to improve the executing efficiency of large scaled deformable collisiondetections is a key problem in computer graphics community. Traditional popularalgorithms usually adapt one or several types of bounding volumes (BV) to create thecorresponding bounding volume hierarchies (BVHs), through which collisiondetections between triangle mesh models are transformed into the intersectiondetections of BVHs, so as to increase the speed of detections. After selecting BVs, itis necessary to build the BVHs data structures, inferring complex geometriccomputation. Although it is usually processed during the pre-computation step, thecost is expensive; in addition, traditional methods based on BVs cannot solve theinner collisions occurring in a geometrically complex object.This paper proposals a novel algorithm executed on image space, avoidingcomplex geometric pre-computation of traditional methods based on BVs. It alsoincludes the SCD (self-collision detection). The algorithm is divided into two parts:one israsterization of mesh models. The colliding objects’ triangle mesh models arerasterized in3D space, obtaining the transform of triangle meshes into rasterization;The other part is computing of minimal volumes of intersection parts betweencolliding objects. Through the rasterization models, the forces applied to each vertexcould be accumulated by forces on each pixel, so as to avoid complex deformationmechanics formulas and to include self-collision detection naturally.The main costs depend on the points of the rasterization and the computation offorces. To solve this issue, we use GPU acceleration in the execution, enablingrasterization processing carried out by several blocks in the same time, so as toimprove the execution speed and efficiency.