Abstract: In view of the current limitations of counting high-density population, a population counting method based on Yolov5 is proposed. The input layer is mainly used for Mosaic data enhancement, adaptive anchor frame and adaptive picture zooming technology. Backbone Yolov5 mainly adopts Focus and CSP( Cross Stage Partial) structure. The Neck layer mainly adopts SPP ( Spatial Pyramid Pooling) module and FPN ( Feature Pyramid Networks)+PAN(Pixel Aggregation Network) structure. The output end adopts CIOU_Loss as the loss function mainly for Bounding Box Loss function and DIOU_Loss as average indicator of the NMS(Non Maximum Suppression). Finally the training results are output. Experimental results show that this method can effectively improve the accuracy of crowd counting.

Key words: people counting, Mosaic data enhancement, Yolov5 algorithm, Focus and cross stage partial (CSP) structure