Robust Deep Learning Method to Detect Face Masks

Abstract

With the outbreak of novel coronavirus (2019-nCoV), wearing masks has become an effective way to prevent the transmission of the virus. But in public places, people are often reluctant to wear face masks and cause the virus to spread widely. This paper uses an efficient and robust object detection algorithm to automatically detect the faces with masks or without masks, making the epidemic prevention work more intelligent. Specifically, we collected an extensive database of 9886 images of people with and without face masks and manually labeled them, then use multi-scale training and image mixup methods to improve YOLOv3, an object detection algorithm, to automatically detect whether a face is wearing a mask. Our experiment results demonstrate that the mean Average Precision (mAP) of the improved YOLOv3 algorithm model reached 86.3%. This work can effectively and automatically detect whether people are wearing masks, which reduces the pressure of deploying human resources for checking masks in public places and has high practical application value.