A 3D Lidar Odometry and Mapping Method Based on Clustering-Directed Feature Point

Lidar odometry (LO) aims to provide robot attitude estimation, which is crucial for robot motion decision-making. In this paper, a new method of lidar odometry and mapping based on clustering-directed feature points (CDP-LOAM) is proposed. The proposed approach introduces a seven-dimensional clustering-directed point (CDP) that encompasses three-dimensional coordinate information, three-dimensional feature vector information, and one-dimensional point cloud category information of LiDAR points. Among them, the point cloud categories are obtained through ground point segmentation and point cloud clustering, and the feature vectors are obtained by processing the classified point clouds with principal component analysis (PCA). The paper proposes a two-step point cloud matching approach and establishes an adaptable weighted comprehensive residual that includes distance and direction residuals of the associated point cloud. The attitude estimation results are obtained by optimizing the residual error. The experimental results demonstrate that this method achieves more accurate positioning results than the traditional lidar method in a wider range of application scenarios.