Probabilistic Matching for 3D Scan Registration

In this paper we consider the problem of three-dimensional scan registration for autonomous mobile vehicles. The problem of 3D scan matching is of enormous importance for the construction of metric representations of the environment, for localization, and for navigation planning in the three-dimensional space. We present a probabilistic technique that computes a probability density for each pair of scans and then performs the registration using a greedy hill-climbing search in the likelihood space. We compare our approach with the ICP-algorithm which currently is the most popular technique for matching 3D range scans. We present experiments illustrating that the estimation error of the ICP algorithm is significantly higher than the resulting error obtained with our probabilistic approach.