Physical Distance and Face Mask Wearing Surveillance System with Deep Learning

Abstract

The COVID-19 pandemic has resulted in the world’s most critical global health catastrophe. To prevent the spread of COVID-19, people are encouraged to maintain 1 meter of physical distance and wear a face mask. However, many people refuse and forget to practice minimum physical distancing and wear their face masks. Besides, manual monitoring of physical distance and wearing face masks are impractical for a large population with insufficient manpower and resources. Hence, this project introduced a physical distance and face mask-wearing surveillance system utilizing deep learning at R&R Malaysia to ensure the safety of travelers during this COVID-19 pandemic. In this project, the system is implemented using the YOLOv4 algorithm to detect masked, non-masked, and incorrect mask-wearing faces and to calculate the physical distance between people. A total of 3,800 custom datasets were prepared to train the face mask detection model. As a result, this model achieved an average mAP of 95.86%, an F1-score of 0.93, and an average loss of 1.3972. The physical distancing detection model is employed on a pre-trained YOLOv4 algorithm to detect people. The Euclidean distance is calculated between the detected bounding boxes to compute the real distance between people.