Robot visual place recognition using COG_SC descriptor

Abstract One of the most active research areas in robotics is visual place recognition using a 3D laser Lidar. A promising research direction involves learning and recognition of scene descriptors, such as scan context descriptors. These descriptors map 3D point clouds to 2D point clouds. Although the scan-context descriptor has a high recognition performance, it is still expensive image data and cannot be handled with low-capacity non-deep models. To address this, we have developed a novel feature descriptor by combining COG (center of gravity) with scan-context descriptor.