A Target-scoring Technique with Anti-shake based on Object Detection and Feature Matching

Training for law enforcement personnel in chestring target shooting is of significant importance, and the completion of such training relies heavily on an accurate and fast target-scoring system. The target-scoring techniques based on digital image processing is low-cost and accurate, and has been improved recently with the development of deep learning. However, the accuracy and robustness cannot be ensured if the target paper shakes or the training environment is complex. To solve these issues, in this paper, we propose an automatic target-scoring method based on feature matching networks and bullet hole recognition models. A feature matching network is first utilized for image registration to reduce the deviations caused by target paper vibrations. Subsequently, a recognition model is employed to detect bullet holes in each frame of the video stream. Finally, a binary tree is established based on the position information of the bullet holes to arrange the sequence of shots and calculate scores, achieving live ammunition target-scoring. Our experimental results demonstrate the effectiveness of our approach, which outperforms existing methods in terms of accuracy, speed, and robustness in target-scoring. This provides insights for further improvements in the target-scoring systems for law enforcement.