Multimodal CNN-Based System For Mask And Maskless Face Detection

Abstract

Face masks existed for much longer before the pandemic corresponding to COVID-19, wherever staff in several sectors, such as medical, chemical, and nuclear, needed to wear masks throughout duties. Following the pandemic caused by the COVID-19 virus, most countries requested publicly covering the nose and mouth as vital life to keep the communities safe. However, 24/7 human superintendence is almost impossible. In this paper, an efficient and automatic multimodal face mask detection was developed. The model was engineered based on intensive investigations, where first, the performance of two well-known deep learning models, particularly MobileNetV2 and VGG19, was investigated. Next, the performance was further improved using the late fusion principle. Four datasets consisting of roughly 6K, 12K, 4k, and 4k images, respectively, are used to confirm the results robustness of the developed model. Overall, the results of the experimental works showed that fusion leads to a more stable and outperforming model compared to five base CNN models, i.e., MobileNetV2, VGG19, and three sequent models.