FDV-YOLO: A Target Detection Network For Dense Pedestrian Detection

Dense pedestrian detection is a hot issue in current research. Due to the detection of different targets, coupled with the interference of occlusion, background complexity and other factors, the current research on dense pedestrian detection has a low accuracy and is prone to leakage or misdetection. To improve the accuracy, this paper proposes an improved model FDV-YOLO based on YOLOv5s, which uses Focal Modulation Network to instead the SPPF layer of the original network, DySample model replaces the original upsampling model, and VariFocalLoss instead of the original CIOU. Experiments on the model on the WiderPerson dataset show that the proposed algorithm improves the precision, recall, and average detection accuracy by 0.6%, 1.1%, and 1.4%, respectively, over the original network. It is verified that the proposed algorithm can effectively improve the accuracy of pedestrian target detection in dense scenes.