Abstract: For solving the problems of poor robustness and threshold- dependence existing ground segmentation methods in different pavement scenes, an accurate segmentation method of Lidar ground point cloud based on LMeds- PCA is proposed. The method constructed a grid model for multi- region segmentation of the ground based on the density of the point cloud and filters out the set of ground seed points. The LMeds- PCA method was used to get the best fitting plane in multiple regions. An accurate segmentation of the ground point cloud of different pavement scenes was realized by determining the confidence level of the fitting planes in each region. The analysis of the measured data processing shows that both the ground segmentation accuracy and recall of the proposed method are maintained at around 95% in different scenes, and the average processing time of 23.7 ms, which has excellent accuracy and real- time performance, meeting the functional requirements of autonomous vehicles.