Research on Visible Light Pedestrian Detection Algorithm Based on Improved YOLOv5m

Visible pedestrian detection technology is widely used in video surveillance, human behavior analysis, intelligent military and other fields. In order to further improve the robustness and detection accuracy of pedestrian detection algorithm, this paper improves on the basis of YOLOv5m algorithm. Firstly, K-means++ clustering is used to optimize the initial center point selection, which makes the algorithm easier to converge in the training process and improves the robustness of the algorithm; Secondly, SENet channel attention module is introduced into YOLOv5m to make the network pay more attention to highlighted targets and improve pedestrian detection accuracy. Finally, OTCBVS data set and INO data set are used for training and testing. The results show that compared with the traditional YOLOv5m algorithm, the proposed improved algorithm has faster convergence and stronger robustness, and the detection accuracy is improved to 93.0%, which can be effectively applied to pedestrian recognition in visible light.