3D Lidar SLAM Based on Ground Segmentation and Scan Context Loop Detection

Improving the accuracy of the vehicle for the simultaneous localization and mapping (SLAM) by reducing the cumulative error has become a hot issue. In this paper, we propose a novel 3D Lidar SLAM method based on ground segmentation and loop detection of global feature descriptor from 3D space (Scan Context). We only use a single sensor (3D lidar) to estimate the 6-DOF poses of the mobile robot. In the preprocessing stage of the 3D lidar point cloud, we complete the global pose constraint using the ground plane segmentation to reduce invalid feature points and the error of feature extraction and matching. Then, the accurate 6-DOF poses are obtained through the lidar odometry and mapping (LOAM), and the global pose is corrected by the Scan Context loop detection method. In the process of global pose optimization, we use a two-stage search algorithm to effectively detect and optimize a loop. Finally, we compare our proposed method with the existing 3D lidar SLAM algorithm on KITTI and our datasets. Experiment results show that the method can effectively reduce the cumulative error. It realizes the robust localization and more accurate map construction based on 3D lidar.