Ray Tracing for natural scenes in parallel processors

Abstract

Ray Tracing is already established as a powerful and elegant technique for image synthesis, but is too inconvenient because of its high computational cost, especially in complex scenes. Because of this, the effort to accelerate Ray Tracing has been important. The utilization of hierarchical bounding volumes produces important time reductions, increasing the memory consumption. Parallelisation in general purpose computers is another way of reducing time. We present an algorithm that permits visualizing by Ray Tracing very complex natural scenes generated by procedural models in distributed memory multiprocessors with any memory size. In this approach, the primitive database is not generated; each processor contains the initial specification of the scene and realizes the visualization, working hierarchically, directly from the specification and a partial generation of the model. The rest of the available memory is used to store some levels of the hierarchical bounding volume, computing in tracing time bounding volumes and primitives not stored. In this way, there are no communications during the tracing phase. The algorithm is also valid for shared memory multiprocessors.