Shape binary patterns: an efficient local descriptor and keypoint detector for point clouds

Abstract Many of the research problems in robot vision involve detection keypoints, areas with salient information input images and generation local descriptors, that encode relevant for such keypoints. Computer solutions have recently relied on Deep Learning techniques, which make extensive use computational capabilities available. In autonomous robots, these are usually limited and, consequently, cannot be processed adequately. For this reason, some tasks still benefit from a more classic approach based keypoint detectors descriptors. 2D images, binary representations visual has shown that, lower requirements, they can obtain performance comparable to real-value techniques. However, achievements not been fully translated 3D where is mainly focused approaches. Thus, paper, we propose detector descriptor patterns. The experimentation demonstrates our proposal competitive against state-of-the-art while its processing performed efficiently.