Abstract: Image-based collision detection algorithms make efficient use of the graphics rendering hardware and reduce the computation overhead of CPU. It can process arbitrary polyhedra, while preserving the merits of image-based collision detection algorithms. This is achieved by decomposing the surfaces of the object into a list of convex pieces. High efficiency is gained by organizing the convex pieces into a hierarchical binary tree and adopting triangle strip compression to accelerate the rendering process. It has been verified by complex benchmarks, giving favorable results in comparison with some related algorithms.