ABSTRACT

In real-time graphics applications, scenes involving cloth motion are very common. Real-time cloth simulation on moving characters is still a challenging task because of the long computation time to solve the physically based cloth model on the CPU. To improve the cloth simulation performance, we propose a level of detail (LOD) method to move part of the cloth simulation work to the Graphics Processing Unit (GPU) on current graphics hardware. Current GPUs are highly parallel stream processors, which execute compute-intensive operations more efficiently than CPUs, but provide less flexible resource access and logic controls. We take advantage of both the two processing units and combine them into a cloth simulation pipeline. The cloth is modeled as a mesh hierarchy. Only the coarse mesh is stored in main memory and processed by the CPU. After the coarse mesh motion is determined by the CPU, it is sent to the GPU for further refinement. We present a set of GPU algorithms that run in multiple rendering passes to subdivide cloth meshes, detect cloth-environment and cloth-cloth collisions and do collision response on the GPU. Our approach can be generalized to handle collision deformation behaviors and can be easily integrated in real-time graphics applications