LSTD: Stable Triangle Descriptor of Line Keypoints for Loop Closing in 3D Lidar SLAM

Loop closing is a crucial module in simultaneous localization and mapping(SLAM), enabling the identification of revisited locations and facilitating trajectory correction. However, the high dimensionality, sparsity, and susceptibility to occlusions of LiDAR-generated point clouds pose significant challenges for this task. We propose Stable Triangle Descriptor of Line Keypoints(LSTD), which extracts texture-rich line features as keypoints and constructs stable triangular features. This descriptor encapsulates both geometric and spatial characteristics, and remains resilient to changes in LiDAR viewing angles, thereby enhancing loop closing capabilities. We devise a voting-based scheme for descriptor matching and leverage the circumscribed circle feature of triangles to enable efficient loop closure detection. Comparative evaluations on the KITTI dataset demonstrate the improvements offered by our approach. Furthermore, we integrate the algorithm into LiDAR SLAM and validate its efficacy on the KITTI dataset as well as our own dataset, showcasing its ability to accurately identify loops and refine the system's trajectory and point cloud map.