Multi-resolution Hierarchical Point Cloud Segmenting

Segmentation is a fundamental issue in point cloud geometry process. It has encountered two difficulties. From one side, those efficient mesh based segmentation algorithms could not be applied to cloud, as point cloud doesn't have topology information; the other difficulty is that the existing point cloud segmentation algorithm could not process large scale models directly. In this paper, we introduce a technique to directly segment a large-scale point cloud into distinct parts . We first construct a simplified approximate geometry model for point cloud based on point cloud's boundary volume hierarchy. This new geometry model is a simplified point model enhanced with topology information. We implement a segmentation algorithm to decompose the simplified model into several parts, and then a new simplified model with better approximation is constructed for each of the segmented parts. Next hierarchy segmentation can be done on these simplified models with better resolution. Our experiment results show that the algorithm is robust and efficient.