A hierarchical model for collision detection of convex polyhedral is presented. This model is based on double representation: the exterior is composed by the minimum outer boxes that envelope the polyhedral; it is an Axis-Aligned Bounding Boxes (AABB), and the interior is composed by the maximum inner boxes that is included inside the polyhedral. Inner boxes allow faster testing of overlapping detection and outer boxes allow faster testing of no-collision. The double representation model proposed has been structured as a binary tree, the polyhedral has been subdivided and the original polyhedral is reconstructed by all nodes of equal level of the tree. Experimental results show that inner boxes improve the collision detection process; consequently, the efficiency of the collision detection is improved. The collision detection algorithm has been implemented and validated with a set of convex polyhedral. Finally, a comparison among the proposed algorithm and some recognized collision detectors has been done.