A G-Octree Based Fast Collision Detection for Large-Scale Particle Systems

Abstract

Particle system has long been a topic of research and has many applications. However, due to its complexity, the fast collision detection between particles and scene geometry is still a bottleneck for real-time or interactive applications. As a fuzzy system, visually correct is more important than physical accuracy at most time for particle system. In this paper, considering its simplicity, we propose to use a simplified version, which can be generated by any simplification method, to approximate the boundary of original scene mesh. We proposed a G-Octree acceleration structure to subdivide the scene space, combining the advantages of both the uniform grid and octree. The resultant G-Octree based agent is uploaded onto GPU to implement a parallel particle system. An encoding scheme is introduced to pack the hierarchical G-Octree structure into a flat 2D texture to exploit the optimized performance of texture memory system on GPU context. Compared with existing methods, the proposed method has potential to handle complex scene in practical applications. The experiments verify its effectiveness.