中国科学院自动化研究所机构知识库

Pre-existing penetrations often show up in many applications, particularly in garments fitting. The popular continuous collision detection (CCD) based methods are incapable of handling them, as there is no history information to rely on. On the other hand, surfaces of human bodies have normals defined to designate their orientation (i.e. front- and back-face), which are totally overlooked by CCD methods (thus they are orientation-free). In this paper we present a history-free method for separating two penetrating meshes, given that one of them represents a rigid object and has clarified surface orientation. This method computes all edge-face (E-F) intersections with discrete collision detection, and identifies illegal vertices with the help of surface orientation, and then builds a number of penetration stencils. On response, the stencil vertices are relocated into a penetrationfree

state, via a global displacement minimizer. The proposed algorithm outperforms existing methods for handling solid/cloth collisions, thus is an effective tool for applications like virtual-try-on and example-based garment animation synthesis.