Face Mask and Social Distance Detection Using Deep Learning Models

Abstract

Coronavirus disease 2019, also known as COVID-19, has become a global threat. In dealing with this global problem of the COVID-19 pandemic, flattening the curve of COVID-19 cases will be difficult unless residents take steps to stem the spread of this virus. The most critical measures during this epidemic are wearing a mask and maintaining social distance in crowded places. This research aims to propose the efficient deep learning model among the few of the most used models in the literature to detect face mask and social distancing violations. The deep learning models MobileNet, You Only Look Once (YOLO), and ResNet-50 are implemented, and their results are compared to ascertain people in a frame and scan for social distance by calculating the Euclidean/Manhattan distance between the bounding boxes generated. Similarly, the proposed model determines if the two people are within 6 feet of the frame; if not, a warning is issued. In this research, Face Mask Lite, Real-World Masked Face, MaskedFace-Net, and Face Mask Detection datasets are considered with 90:10, 80:20, 70:30, and 60:40 ratios to train and test these deep learning models. These deep learning models are processed in Anaconda + Python 3.x (3.8 or older) environment. The proposed model can be useful in hospitals and clinics, airports, shops and workplaces, and other crowded places where social distancing and face masks are required to avert the spread of not only COVID-19 but also other communicable diseases as well.