A Triangular Descriptor with Distance Information for Loop Closure Detection in LiDAR SLAM

In the field of mobile robotics, loop closure detection serves as an essential component of Simultaneous Localization and Mapping (SLAM). Compared with vision-based approaches, LiDAR-based loop closure detection offers more stable geometric features. This paper proposes a novel triangular descriptor with distance information. The proposed method extracts keypoints through adaptive Euclidean clustering, constructs distance vectors based on keypoint relationships, and generates the descriptor by integrating triangular geometric features. Experiments on the KITTI dataset demonstrate that our method outperforms existing approaches with superior maximum F1 scores across multiple sequences. Integration of the algorithm into the LeGO-LOAM system reduces cumulative drift error by 31.16%, validating the effectiveness of the proposed approach.