Weighted Scan Context: Global Descriptor with Sparse Height Feature for Loop Closure Detection

Abstract

In simultaneous localization and mapping (SLAM), the loop detection module has an important impact on the global consistency of trajectory and mapping. Compared to vision-based methods, LIDAR-based methods are gaining attention because they can acquire a complete 360° horizontal field of view and are not affected by changes in illumination. However, pure height features extracted from LIDAR point clouds are subject to feature degradation in specific environments. To reduce the adverse effects of height feature degradation, we propose a global descriptor, Weighted Scan Context, that uses the intensity information of the points to sparse the geometric features in the height direction. To reduce the sensitivity of cosine distance to viewpoint translation motion, a hybrid distance metric that integrates cosine distance and Euclidean distance is used to measure the similarity of two scenes. Through experiments on the KITTI dataset, the proposed method shows better performance compared to existing methods.