Hierarchical OBB-sphere tree for large-scale range data management

Recent research has shown that range data acquired from consumer imaging devices are successfully used to build accurate 3D geometric models of surrounding environments very quickly. However, as the volume of modeling space grows, its processing time inevitably becomes huge when a simple linear data structure is used to handle a large range data set. The main difficulty is that pairwise scan matching among all pairs of input data requires quadratic computational complexity. The OBB-sphere tree presented in this paper accelerates the process by intelligently choosing a subset of related pair candidates. The proposed data structure is an enhanced sphere tree in which every leaf node contains a set of 3D points computed from a depth image and bounded by one sphere and one tighter-fitting shape, which is the oriented bounding box (OBB). This approach exploits the simplicity of spheres in hierarchical tree construction and the intersection test which is performed to reject objects that are far apart. The compactness of OBBs is used to refine the results of the intersection test. The number of possible pairs was reduced by more than half in some test datasets in the experiment.