Combining two point clouds generated from depth camera

Abstract

Depth Cameras are widely used in mobile robotics recently. Combining two frames is the key step to construct a complete 3D map, and the relative pose between two views w.r.t two frames is needed. 3D point pairs are necessary for computing 6DOF pose transformation. In this paper, a method for finding correspondences of two point clouds is proposed, and the advantages of depth camera are taken fully. The idea of this approach comes from the fact that people pay more attentions to some key points when they watch an environment. Point pairs are generated from just points around the feature points, and these pairs are corresponded accurately and the number of pairs is sufficient to compute relative pose. Firstly, the algorithm for matching features from intensity image is proposed. Secondly, 3D point pairs are obtained according to the feature positions. Lastly, two point clouds are registered based on the relative pose computed. The experiments demonstrate the efficiency and the effectiveness of the approach.