The problem of relative pose measurement of non-cooperative targets in space becomes the most difficult task of the in-orbit operation task. By clustering the three-dimensional point cloud of the target obtained by Lidar, a small-scale cluster point cloud with obvious characteristics is obtained, which effectively improves the registration efficiency and precision. Aiming at the problem that clustering algorithm based on region growth can not recognize clusters with similar features when clustering visible point clouds, a method for optimizing 3D point cloud clustering with 2D images was proposed. This method established a mathematical mapping relationship between the depth value information and the RGB color value. After the point cloud dimension was reduced, the boundary was extracted using the color gradient mutation, the points within the boundary were reversely restored to the original point cloud, and finally the various types of point clouds were combined to obtain a point cloud with distinctive features that were easy to recognize. The experimental results show that under the condition of the registration angle error is ±5°, the size of the point cloud can be effectively reduced and the significant features are preserved. It provide technical support and solutions for the real-time measurement of relative pose of non cooperative targets in space.