Since the vehicle laser point cloud data usually contain a large number of points, including noise points, and directly calculate the adjacent points of each point, the road surface point cloud extraction is not accurate, resulting in low detection accuracy. Therefore, a pothole detection method based on vehicle-mounted laser point cloud data was studied. Firstly, the KD tree algorithm and the CFS algorithm optimized by least square method are used to filter the laser point cloud data of the vehicle, avoid directly calculating the adjacent points of each point, reduce noise and improve fitting accuracy, so as to accurately extract the road surface point cloud, and map the road surface point cloud to the XOY plane through mapping processing to generate road surface images. The low frequency part of the image is processed by gamma correction, the high frequency part of the image is processed by bilateral filtering, and the enhanced image is obtained by fusion of the above processing results. The pothole area is determined by the characteristics of the connected domain, and the pothole depth is calculated to realize the pothole detection. The experimental results show that the proposed method can effectively extract the pavement point cloud, the pavement image processing effect is good, and 5 pavement potholes are accurately detected, and the shape of the detected potholes is basically consistent with the actual shape, and the position deviation is small, which has a good image processing effect and accurate detection result.