Automatic 3D point cloud registration based on hierarchical block global search

A improved Iterative Closest Point(ICP) algorithm based on hierarchical block global search to neighbor local search method is presented to get up the registration speed of the ICP algorithm and remove the effect of defective point clouds on the point cloud registration. The method aims at finding the corresponding closest points for ICP algorithm and resulting in the automatic registration of 3D point clouds. After the initial registration, merely a few model points are selected hierarchically while the point cloud blocks are served as the selection units. Then, the corresponding closest points of those model points are searched globally. After a large number of neighboring points of a few model points are selected, the corresponding closest points of the vast number of the model points are searched in local areas by considering the closest points of the few model points as the searching centers. Finally, the correspondence outliers are removed, and the fine alignment transformation is obtained. As compared to both the traditional ICP algorithms based on KD-Tree and LS+HS(Logarithmic Search Combined with Hierarchical Model Point Selection), the proposed algorithm has improved its registration speeds by 78% and by 24% for the Happy bunny scanning data as well by 73% and by 30% for Dragon scanning data. It concludes that the proposed algorithm can quickly and precisely achieve the registration of 3D point clouds.