Spatial object median rules in bounding-volume hierarchies for complex environments

Abstract

Splitting process in Bounding-Volume Hierarchies (BVH) for collision detection is one of the most challenging issues in computer graphics. The splitting process requires an object with their set of triangles to be splitted into two parts using binary tree. The problem exists where most of objects do not have similar triangle size and thus creating unbalanced tree when reaching the final level of tree construction. It is very crucial to make sure that the BVH tree construction is always in balanced as the speed of BVH tree traversal algorithm dropped for unbalanced tree. In this thesis, we introduce the Spatial Object Median Splitting (SOMS) Rule to enhance the capability of BVH construction by efficiently splitting the irregular triangles so that each of them can be bounded with single Axis-Aligned Bounding Box (AABB). The process starts by splitting the longest axis between the midpoint of the triangles based on their minimum and maximum points. Result show that SOMS is capable of creating an optimum level of BVH where more leaf nodes containing a single triangle bounded with AABB are produced compared to Spatial Median technique. From the BVH construction experiments, SOMS managed to create the AABB tree in 43 milliseconds for object that have 948 triangles or 8% faster compared to the Spatial Median technique. Furthermore, experiment to create BVH also showed that SOMS produced 26% more nodes with single triangle compared to the Spatial Median technique given that the total object triangle count for the complex environment are 5672 triangles. The E-Node traversal approach that is specially designed to perform traversal testing for SOMS technique is able to detect overlap between dynamic and static models in an average of 0.2 milliseconds for the same complex environment. As a conclusion, BVH can easily be constructed using SOMS approach to create a more balanced tree and an E-Node traversal algorithm that is more efficient and faster.