Aiming at the problem of automatic recognition of target points in locomotive dynamic bounding images, a target point recognition algorithm based on dual-resolution analysis is proposed to solve the problem of searching efficiency for multi-target in high-resolution images. A self-describing vector is proposed to improve the traditional RANSAC algorithm to reduce the interference of background noise. The homography matrix of different target image points is solved. Finally, efficient and accurate automatic recognition and matching of target points can be achieved.