Abstract: In simultaneous localization and mapping (SLAM), loop detection is an important and challenging problem. Existing point cloud-based loop detection usually only uses height information to construct local or global descriptors, but its single description ability leads to many false loop detections. In order to solve this problem, combining the height, intensity and density information, an improved global descriptor is proposed. In order to use this global descriptor for efficient loop detection, several loop candidate is searched through K-Dimensional Tree (KD-Tree) firstly, and then the idea of Hamming distance is used to calculate the distance between two frame point clouds corresponding to the global descriptor. Finally, determine the frame with the highest similarity as the loop frame. The experiment uses public data sets to test the performance of the algorithm. The results show that, compared with the comparison algorithm, the algorithm has the higher precision and recall rate. At the same time, the SLAM framework integration experiment is carried out to obtain superior positioning and mapping accuracy.