Fast Relocalization and Loop Closing in Keyframe-Based 3D LiDAR SLAM

Relocalization and Loop closing play important roles in robotic simultaneous localization and mapping (SLAM). In this work, we propose a light-weight bag of words (BoW) method for the relocalization and loop closing in 3D LiDAR SLAM, which is based on BoW3D. Different from original BoW3D, to further improve the retrieval efficiency, we use the words of keyframes to build the vocabulary of the database. Based on the light-weight database, we propose a relocalization method to deal with the tracking failure cases, and also propose a keyframe-based loop closing method to improve the accuracy of 3D LiDAR SLAM. In experiment, we test the recognition performance of keyframe-based database, and evaluate the relocalization performance. To evaluate the loop closing performance in practice, we furthermore embed the keyframe-based BoW3D to LiDAR odometry system. The experimental results show that the proposed place recognition method not only achieves competitive retrieval performance but also shows great recognition efficiency.