A LiDAR SLAM based on multi-feature fusion for unstructured environments

Traditional LiDAR Simultaneous Localization And Mapping (SLAM) algorithms typically only extract geometric features of corners and planes, ignoring other types of geometric features such as vertices, pillars, and facades, as well as intensity features, resulting in lower accuracy in unstructured environments. To solve this problem, we propose a new LiDAR SLAM framework that can extract multiple types of geometric and intensity features of the environment and fuse them in a coarse to fine manner to improve localization accuracy. Firstly, we use ground filtering to separate the scan into ground and non-ground points. Then, Principal Component Analysis (PCA) is used to roughly extract multiple features on non-grounded points, and curvature is used for fine feature analysis on all point clouds. Finally, we perform weighted fusion on multiple features extracted by rough and fine methods to obtain more accurate feature extraction results. Experiments on public datasets have shown that our proposed method has higher accuracy compared to pure geometric LiDAR SLAM systems.