Abstract: Aiming at the problem of lack of credible feature points in large point cloud terrain mapping of ground 3D scanner, a method flow of high-precision automatic positioning and extraction of spherical target center is proposed. The point cloud intensity map is obtained from the 3D point cloud, and the intensity map is divided into multiple subgraphs. The candidate target region of interest(ROI) in the subgraph is obtained by using the existing image processing algorithm. An improved Zernike moment sub-pixel edge detection algorithm based on one-dimensional maximum entropy is proposed to accurately obtain the pixel set of candidate targets in each ROI. The credible target is determined by the relevant threshold of the target radius and the sphere rate.The point cloud of the trusted target is alternately performed to fit the ball and eliminate gross errors,and the accurate target center is obtained. The experimental results show that when the measurement range of the target ball with a diameter of 14.5 cm is within 80 m, the success rate of the algorithm to extract the target from the point cloud is 96.7%.Based on the method of automatically extracting the center of the spherical target,the scanning coordinate system is converted to the specified coordinate system,and the mean square error of the three items(elevation mean square error/plane mean square error/3D space error)is less than 1 cm.

Key words: large point cloud, spherical target, 3D scanner, Zernike, target extraction