1 - Zhejiang University, China

2 - University of North Carolina at Chapel Hill, USA

3 - Alibaba-Zhejiang University Joint Institute of Frontier Technologies, China

Benchmarks: Our novel GPU-based collision detection algorithm is used to simulate clothes with regular (a-c) or irregular (d-f) shapes, which may correspond to a single layer (b, e) or multiple layers (a, c, d, f). We use a parallel spatial hashing algorithm to perform inter-object and intra-object collision checking on a GPU along with normal cone culling. The yellow line segments highlight all the areas that have no self-collisions and can be culled away. As compared to prior GPU-based collision detection algorithms, we observe 6−8X speedup.

Abstract

We present a GPU-based self-collision culling method (PSCC) based on a combination of normal cone culling and spatial hashing techniques. We first describe a normal cone test front (NCTF) based parallel algorithm that maps well to GPU architectures. We use sprouting and shrinking operators to maintain compact NCTFs. Moreover, we use the NCTF nodes to efficient build an enhanced spatial hashing for triangles meshes and use that for inter-object and intra-object collisions. Compared with conventional spatial hashing, our approach provides higher culling efficiency and reduces the cost of narrow phrase culling. As compared to prior GPU-based parallel collision detection algorithm, our approach demonstrates 6?8X speedup. We also present an efficient approach for GPU-based cloth simulation based on PSCC. In practice, our GPU-based cloth simulation takes about one second per frame on complex scenes with tens or hundreds of thousands of triangles, and is about 4-6X faster than prior GPU-based simulation algorithms.
 

Contents