Fast collision detection between high resolution polygonal models

Abstract

In this paper, we address the problem of collision detection between very high resolution polygonal models. We propose an approach that uses the EBP-Octree data structure in a single-core CPU architecture. The algorithm manages a set of files to support a cache-like out-of-core algorithm that performs the collision and intersection tests. Our method uses a data structure called EBP-Octree (Extended Bounding-Planes Octree), which is a very tight hierarchy of convex bounding volumes that, based on a spatial decomposition of the model using an octree, defines a bounding volume at each node by a subset of the planes of the portion of polygonal model contained at that node. The system adapts the memory consumption to the system’s hardware features so it can provide exact triangle collision detection in almost any environment, since the amount of information in the main memory can be limited just varying the threshold of the EBP-Octree. We have tested the data structure and the algorithm in scenarios where two models of 28 million polygons move through one another, achieving real time frame rates for first collision tests on a commonly available computer.