Abstract: In this paper, we propose a feature extraction and matching method based on point cloud segmentation to address the problems of information redundancy and interference of discrete points in laser odometry algorithms. By utilizing the horizontal rotation scanning characteristics of mechanical LiDAR, the point cloud data is segmented and clustered by scan lines, and edge points are extracted as line features and surface points are extracted as surface features. Compared to traditional feature extraction methods, our algorithm can effectively extract feature points with less distinctiveness and eliminate discrete points. The algorithm not only reduces the computational complexity but also improves the accuracy and robustness of positioning, which has great application prospects in robot navigation tasks. The performance of the algorithm is validated through computer simulation and achieved good experimental results.