A LiDAR Loop Closure Detection Algorithm Integrating Multiple Information of Point Cloud

In SLAM (Simultaneous Localization and Mapping) systems, the odometry in the front-end performs inter-frame matching in unstructured environments. However, due to the accumulation of errors, the accuracy of localization and mapping deteriorates. Without external observations, the loop closure optimization strategy triggered by the loop closure detection algorithm is a key method for reducing cumulative errors. Therefore, a LiDAR SLAM loop closure detection algorithm integrating multiple information of point cloud is designed. The algorithm focuses on encoding the height and density information of point cloud. Firstly, a kd-tree-based fast screening of loop closure frames is performed. Then, a refined loop closure frame selection based on similarity calculation is conducted, followed by a geometric consistency verification based on the NICP method. Experimental results on an online unstructured dataset demonstrate that the root mean square error (RMSE) of the absolute trajectory error after optimization by the proposed algorithm is reduced to 0.87 meters, showing a performance improvement of approximately 36%, with point cloud mapping results being more accurate and detailed. Moreover, the precision-recall (PR) curve of the proposed algorithm covers the largest area compared to other loop closure detection algorithms such as Scan Context and Iris.