The collision detection in virtual scene can effectively improve the fidelity of virtual reality, however, the existing collision detection algorithm are not fully ideal. In this paper, a collision detection optimization algorithm based on hybrid bounding volumes is presented to improve the real-time ability of bounding volumes collision detection. The algorithm combines the simplicity of AABB-AABB overlap test with the tightness of k-dops. Bounding volumes binary trees of the objects are designed as double-layered structure, which employ AABBs at the top layer, and k-dops at other layers. The intersection test uses different methods to detect the contact status among objects. Quest tree method is used to speed up the algorithm. Experiment results show that the algorithm above is efficient in improving the real-time and accuracy of collision detecting.