Partitioning and Handling Massive Models for Interactive Collision Detection



We describe an approach for interactive collision detection and proximity computations on massive models composed of millions of geometric primitives. We address issues related to interactive data access and processing in a large geometric database, which may not fit into main memory of typical desktop workstations or computers. We present a new algorithm using overlap graphs for localizing the “regions of interest” within a massive model, thereby reducing runtime memory requirements. The overlap graph is computed off-line, pre-processed using graph partitioning algorithms, and modified on the fly as needed. At run time, we traverse localized sub-graphs to check the corresponding geometry for proximity and pre-fetch geometry and auxiliary data structures. To perform interactive proximity queries, we use bounding-volume hierarchies and take advantage of spatial and temporal coherence. Based on the proposed algorithms, we have developed a system called IMMPACT and used it for interaction with a CAD model of a power plant consisting of over 15 million triangles. We are able to perform a number of proximity queries in real-time on such a model. In terms of model complexity and application to large models, we have improved the performance of interactive collision detection and proximity computation algorithms by an order of magnitude.