The traditional collision detection algorithm based on the point cloud typically uses bounding volume hierarchies or space decomposition to determine whether there is collision. This method cannot achieve the accurate safety distance value between the object and the scene. In this study, a collision analysis method based on the binocular stereo point cloud is proposed, which is mainly for nonconvex complex surface objects. Firstly, the binocular stereo algorithm is used to reconstruct the point cloud of the scene captured by the calibrated camera. Then, the point clouds of the object and the scene are both utilized to solve the collision problem. The process distance values are rapidly obtained by the K-D tree search algorithm. The symbol is defined by the coordinate relationship of the point cloud along the optical axis of the camera. The accuracy of this method in the field experiment of a detector is 100% . Compared with the existing algorithms, this method can obtain the distance of each point on the surface of the object. The complexity of calculation is reduced efficiently under the reference of camera′s optical axis. The detection time at the drop sampling single point is not larger than 0. 15 s, which can satisfy the need of rapid collision analysis between non-convex objects and complex topography. This method can successfully complete the sampler-terrain collision analysis task of sampling point selection in the lunar surface sampling package of Chang′e-5.