Research on Head and Shoulders Detection Algorithm in Complex Scene Based on YOLOv5

For subway stations and other places with dense traffic, the effect of pedestrian detection is easily affected by occlusion. At the same time, the head and shoulder detection has strong anti occlusion ability and low computational requirements. In view of the existing problems, this paper proposes a head and shoulder detection algorithm based on yolov5 in complex scenes, and designs and adds a deformable convolution network and attention mechanism module in the feature extraction network to improve the ability of feature extraction. The experiment is designed to train and test the pedestrian head and shoulder data set in the subway complex environment. The experimental results show that the accuracy of pedestrian head and shoulder detection for small target is improved in subway complex scene.