A fast spatial partition method in bounding volume hierarchy

Acceleration data structure plays a critical role in ray tracing. As a popular acceleration data structure, bounding volume hierarchy(BVH) is preferred due to its low memory cost and flexibility. The efficiency of BVH can be affected by the non-uniform scene. This paper presents a novel algorithm which is able to build a tree of high quality and high building efficiency. The algorithm is based on the depth attribute of BVH tree, and it takes different building strategies: utilizing spatial partition in shallow levels, while using object partition in deep levels. It is difficult to get an optimal spatial partition plane in shallow levels, since it is a tradeoff between building efficiency and tree quality. The object partition method is used under some assumption to find an appropriate optimal solution. The experiment results indicate that our algorithm takes about 39% of the building time to build the similar quality BVH tree compared with the latest algorithm Spatial Splits in Bounding Volume Hierarchies (SBVH) on average.