AbstractImage‐based collision detection algorithms make efficient use of the graphics rendering hardware and reduce the computational cost of CPU. In this article, a fast collision detection algorithm based on image space is presented, which combines graphics hardware capabilities with a simplified geometric representation (oriented bounding box) in order to rapidly detect collisions between complex objects. The method can deal with arbitrary polyhedra, while preserving the merits of image‐based collision detection algorithms. This is achieved by decomposing the surfaces of an object into a list of convex pieces. High efficiency of the algorithm is obtained by organizing the convex pieces into a binary tree with each node representing a convex piece, and by adopting triangle strip compression. The algorithm has been implemented and compared with related algorithms. Copyright © 2003 John Wiley & Sons, Ltd.