Improvement of Human Recognition Algorithm in Public Places Based on YOLOv5

In the field of human recognition in video surveillance, this paper proposes a higher detection accuracy object detection algorithm YOLO-GAM+to address issues such as dense object occlusion and low detection rate of small targets. By adding a layer of GAMAttention module to the Backbone of yolov5, and then adding a detection layer to the head, the detection accuracy of the model was effectively improved. Finally, by introducing soft nms, the detection accuracy of the model was further improved. The experiment showed that the mAP50 of YOLO-GAM+ reached 76.8%, which increased by 2.1 percentage points while the parameter quantity remained basically unchanged. Experimental results have shown that the improved algorithm proposed in this article is more suitable for the deployment of human recognition in video surveillance.