Automatic detection of face mask wearing based on polarization imaging

Abstract

Amidst the global health crisis sparked by the coronavirus pandemic, the proliferation of respiratory illnesses has captured worldwide attention. An increasing number of individuals wear masks to mitigate the risk of viral transmission. This trend has posed a critical challenge for the development of automatic face mask wearing detection systems. In response, this paper proposed what we believe is a novel face mask wearing detection framework DOLP-YOLOv5, which innovatively employs polarization imaging to enhance the detection of face mask by leveraging the unique characteristics of mask surfaces. For extracting essential semantic details of masks and diminish the impact of background noise, the lightweight shuffle attention (SA) mechanism is integrated in the backbone. Further, a Content-Aware Bidirectional Feature Pyramid Network (CA-BiFPN) is applied for feature fusion, sufficiently integrating the information at each stage and improving the ability of the feature presentation. Moreover, Focal-EIoU loss is utilized for the bounding box regression to improve the accuracy and efficiency of detection. Benchmark evaluation is performed on the self-constructed polarization face mask (PFM) dataset compared with five other mainstream algorithms. The mAP50-95 of DOLP-YOLOv5 reached 63.5%, with 3.08% and 4.44% improvements over the YOLOv8s and YOLOv9s, and achieved a response speed of 384.6f/s. This research not only demonstrates the superiority of DOLP-YOLOv5 in face mask wearing detection, but also has certain reference significance for other detection of polarization imaging.