Face Mask Detection Using Deep Learning For Public Places

Over the past century, before and after the pandemic of COVID-19, workers in multiple sectors, such as medical, chemical, and nuclear, have been required to wear face masks during duties. However, physical 24/7 supervision is nearly impossible in public places. With the outstanding performance achieved by deep learning almost in all fields, this problem can be easily solved by building an automated mask detection system. This paper investigates the performance of five deep learning models, particularly Convolutional neural networks(MobileNetV2, VGG19, and three sequential models) when used for mask detection, i.e., automatically distinguishing between a person wearing a face mask and a person who is not. To ensure the results robustness of this comparison, four datasets consisting of approximately 6K, 12K, 4k, and 4k  images, respectively, have been used.  Overall, the results of the experimental works showed that all models achieved a good performance when processing the first, second, and fourth datasets, with some improvement achieved by both MobileNetV2 and VGG19. However, when processing the third dataset containing low-quality images, MobileNetV2 significantly outperformed others.