Mask recognition system based on anchor

During the epidemic prevention and control phase, masks must be worn in public places according to the epidemic prevention regulations. However, manual detection may result in missed detections when there is a high flow of people. Therefore, using computers to complement manual detection can reduce costs and improve efficiency. For the recognition system of mask-wearing situation, this system utilizes the target detection lightweight model; the difference between the two is that the former model is optimized for mobile, which is more suitable for deployment on devices with limited computing power such as mobile or edge detection. And the model is based on the backbone network of FaceBoxes face detector to complete the detection of wearing masks. After the experimental demonstration results, only 5 sample results of the 50 face images we tested (including 23 photos with masks and 27 without masks) showed errors. The system had a 90% correct rate in natural scenes throughout the day.

I. Introduction

Currently, the new crown epidemic is ravaging the world. When entering public places, epidemic prevention regulations require people to wear masks. Still, the task of implementing that everyone wears a show is so large that it cannot be accomplished manually alone. Therefore, we developed a mask-wearing monitoring system to make it easier, more accurate, and more extensive to monitor people wearing masks in public places.