In the optomechanical design of laser radar, the light source center of lidar does not coincide with the mechanical rotation center because the parameters need to meet the requirements of the actual project, which will lead to the bending of point cloud image and seriously affect the performance of lidar. Therefore, combined with the actual scanning situation of the three-dimensional radar, first, the reason for the bending of the lidar point cloud image is analyzed and the error calculation formula for each scan line is derived. Then, by modifying the parameters, the pixels in the relative coordinate system are transformed into the depth values in the world coordinate system, so as to complete the transformation of the coordinate system. Finally, the bending of the point cloud image is corrected. At the same time, the reliability of the modified algorithm is analyzed in detail combined with five point cloud images. Experimental results show that after correcting the algorithm, the point cloud image is flat everywhere, almost no bending, the contour of all test obstacles has no obvious deformation, the pixel points are arranged orderly, and the ranging error is reduced from 10.2 cm to less than 2 cm. In addition, the lateral range resolution of the nearest （0,1.8 m） and the longitudinal range resolution of the radar in the monitoring area are 7.5 cm and 4.8 cm, respectively, and the lateral range and longitudinal range resolution of the farthest point （25 m, 4.5 m） are 20 cm and 6.1 cm, respectively. Compared with the traditional data matching and splicing model, it is proved that the proposed algorithm of coordinate system transformation can fundamentally solve the two kinds of nonlinear errors caused by the misalignment of coordinate centers and the rotation of galvanometer. Moreover, it is proved that the algorithm has high stability through the test experiments in snow and complex environment.