Fast Point Feature Histograms (FPFH) for 3D Registration.pdf

Fast Point Feature Histograms (FPFH) for 3D Registration Radu Bogdan Rusu, Nico Blodow, Michael Beetz Intelligent Autonomous Systems, Technische Universit ¨at M ¨unchen {rusu,blodow,beetz}***@ Abstract—In our recent work [1], [2], we proposed Point Feature Histograms (PFH) as robust multi-dimensional features which describe the local geometry around a pointpfor 3D point cloud datasets. In this paper, we modify their mathematical expressions and perform a rigorous analysis on their robustness plexity for the problem of 3D registration for overlap- ping point cloud views. More concretely, we present several optimizations that reduce putation times drastically by either caching puted values or by revising their theoretical formulations. The latter results in a new type of local features, called Fast Point Feature Histograms (FPFH), which retain most of the discriminative power of the PFH. Moreover, we propose an algorithm for the putation of FPFH features for realtime applications. To validate our results we demonstrate their ef?ciency for 3D registration and propose a new sample consensus based method for bringing two datasets into the convergence basin of a local non-linear optimizer: SAC-IA (SAmple Consensus Initial Alignment). I. I NTRODUCTION In this paper we tackle the problem of consistently aligning various overlapping 3D point cloud data views into - plete model (i