PBS-LeGO-SLAM: Lightweight, High-Precision and Robust LiDAR Odometry and Mapping for Complex Environments

3D LiDAR SLAM technology is essential for autonomous driving and mobile robots, but current techniques face issues like z-axis trajectory drift and inefficient loop closure detection in complex environments. There's a need for a balance between real-time performance, accuracy, robustness, and computational efficiency. This paper introduces PBS-LEGO-SLAM, a lightweight, high-precision, and robust framework based on LeGO-LOAM. It utilizes several advanced techniques for 3D LiDAR SLAM. It employs the Patchwork++ algorithm for ground segmentation, after projecting the 3D point cloud into a range image, the points are labeled as ground or non-ground. Ground points are then projected to form the ground image. Edge and planar features are extracted from the projection image based on pixel smoothness, and LinK3D descriptors are created for matching 3D point features using a keypoint aggregation algorithm. Ground-optimized odometry and mapping algorithms estimate the pose, with the BoW3dalgorithm detecting local loops and updating the local map in real-time, optimizing the 6-DoF odometry pose. Finally, the Scan Context descriptors are used for global loop closure detection. Experiments on the KITTI dataset show that PBS-LEGO-SLAM improves accuracy and robustness while meeting real-time and lightweight computational requirements compared to baseline methods.