3D Point Set Registration based on Hierarchical Descriptors

Registering partial point clouds is crucial in numerous applications the field of robotics, vision, and graphics. For arbitrary configurations, registration problem requires an initial global alignment, which computationally expensive often still refinement. In this paper, we propose a pair-wise method that combines fast convergence made possible by hierarchical surface descriptors with arbitrarily fine sampling enabled continuous representations. Registration performed matching increasing resolution – surfaces allow us to choose high while restricting search space according hierarchy. We evaluated our on large set problems, demonstrating very competitive accuracy makes subsequent refinement local unnecessary.