Coherent hierarchical collision detection for clothing animation

Abstract

Collision detection has been one of the major performance bottlenecks in clothing animation. Currently the mostly used approach uses the bounding volume hierarchy (BVH). In order to further accelerate the collision detection speed, this paper proposes a new coherent hierarchical collision detection method. This method takes advantage of the fact that in a clothing animation system, the clothing is generally adjacent to the mannequin when it is dressed, resulting in the two BVH trees for the two models to intersect with near probability one in the upper half layers, thus causing many unnecessary calculations during collision detection. Based on this observation, this paper proposes a collision detection structure with only one axis-aligned bounding box (AABB) tree for the mannequin and builds an adaptively updated coherent relationship between each clothing vertex and its nearby AABB nodes of the mannequin. With the proposed coherent hierarchical collision detection method, the overall animation performance is enhanced more than twice over the traditional BVH approach. Furthermore, the proposed method is very suitable for parallel implementation, and our simulation using CUDA shows a further 40 times speed-up.