LiDAR SLAM method in outdoor environment based on global descriptor loop detection

Abstract In outdoor environments, the odometer based on lidar sensors has accumulated errors, which may lead to map drift and map construction failure. This paper proposes a SLAM method based on global descriptor loop detection. This method uses stable triangle (STD) descriptors for loop closure detection to reduce cumulative errors. At the same time, this paper also proposes a candidate loopback frame search algorithm based on inverse document frequency (IDF) to ensure that high query efficiency can be maintained when the number of descriptors increases. In addition, the iterative closest point method (ICP) weighted by multi-scale residual terms is used to obtain the pose transformation between matching loop closure frames, thereby effectively improving the accuracy and speed of loop closure detection. To verify the performance of the proposed algorithm, we conducted experiments on the public dataset KITTI. Experimental results show that compared with existing algorithms such as LeGO-LOAM and Scan-Context, the algorithm proposed in this article has better performance in outdoor scenes and can improve the accuracy of pose estimation and mapping accuracy.