Mask wearing detection algorithm based on improved YOLOv5

Abstract

The new coronavirus can be transmitted through respiratory droplets and other means. By properly wearing a mask, the transmission of the virus can effectively be prevented. A mask wearing detection method based on improved YOLOv5 is proposed to address the problem of low accuracy of mask detection due to complex factors current mask detection algorithms, such as occlusion, dense crowds and small-scale targets. By combining with a bi-directional feature pyramid network to improve the neck backbone networks, the fusion of high-resolution and low-resolution features is enhanced to improve the accuracy of small-scale target detection. Landmarks, a key point feature for faces, is also introduced into the detection layer of YOLOv5, which improves the detection accuracy of occlusion and dense crowds. The experimental results show that the algorithm can meet the requirements of accuracy and real timeliness of the mask wearing detection under conditions with limited resources.