Comparison of nearest-neighbor-search strategies and implementations for efficient shape registration

Abstract

The Iterative Closest Point (ICP) Algorithm is one of the most popular approaches to shape registration currently in use. At the core of ICP is the computationally-intensive determination of nearest neighbors (NN). As of now there has been no comprehensive analysis of competing search strategies for NN. This paper compares several libraries for nearest neighbor search (NNS) on both simulated and real data with a focus on shape registration. In addition, we present a novel efficient implementation of NNS via k-d trees as well as a novel algorithm for NNS in octrees.