Research on face detection based on improved YOLOv5s in dense crowd scenes

With the acceleration of urbanization and population growth, face detection in densely crowded scenarios has become increasingly important. To balance the detection speed and detection accuracy, this thesis proposes a face detection method based on YOLOv5s, which aims to solve the problems of missed detection and low recall in dense crowd environments. In this paper, K-means++ clustering is used instead of Kmean clustering to adjust the anchor frame size of the YOLOv5 algorithm, and the global context block and the C3 module are fused to form the C3GC module to enhance the feature extraction of the model for faces in dense crowds. Meanwhile, a new loss function calculation method, named KF-EiOU, is proposed in the article to speed up the convergence of the model. The final experimental results show that compared with the original YOLOv5s, the mAP value of the improved model is increased by 3.46%. The improved model has better generalization robustness and accomplishes better face detection in dense crowd scenes.