Research on the application of computer vision in the police equipment : Auto-targeting through YOLOv5s as an example
Ruixi Liu, Wanqiu Zhang, Yuxin Xiao, Yinan Yang · 2022 7th International Conference on Intelligent Computing and Signal Processing (ICSP) · 2022
The previous automatic targeting techniques are prone to deviation and low robustness. To address the problem that physiological factors are liable to misfire and miscalculation when police shoot in critical situations, an automatic target recognition assistance system based on a lightweight and efficient YOLOv5s network is proposed. The system first introduces a lightweight feature-enhanced representation module to detect and classify targets according to real-time images captured by RGB cameras, which reduces the false detection rate of the network. Then, the user selects a controller based on the system's labeling of the target Head and Cavity. Finally, the feature-enhanced representation module extracts higher resolution maps, which helps in tiny target detection. Experiments show that the system maintains the lightness and efficiency of the YOLOv5s network. It achieves an average detection accuracy of 83.3% for Head and Cavity, which meets the performance requirements of AI models in police application scenarios and provides a theoretical basis for future computer vision applications in police equipment development.