Research on Mask Wearing Detection of Natural Population Based on Improved YOLOv4

Abstract

Recently, the domestic COVID-19 epidemic situation has been serious. At present, the most effective epidemic prevention and control measure is still to wear masks. Therefore, setting up automatic detection devices for wearing masks in public places can better help relevant departments to carry out epidemic prevention and control work. Aiming at the problems of the existing mask detection algorithms, such as low accuracy, poor robustness, and inability to meet the real-time requirements of the proposed method, this paper proposes a new mask wearing detection method based on improved YOLOv4. Specifically, first of all, we add the Coordinate Attention Module to the backbone to coordinate feature fusion and representation. Secondly, we conduct a huge number of network structural improvements to enhance the model's performance and robustness. Thirdly, we deploy the K-means clustering algorithm to make the nine anchor boxes more suitable for our NPMD dataset. The extensive experimental results show that the improved YOLOv4 performs better, exceeding the baseline by 4.06% AP with a comparable speed of 64.37 FPS. It can complete a comprehensive and accurate mask wearing detection task in natural scenes.