Social Distancing and Face Mask Detection using Deep Learning Models: A Survey

Abstract

Over the recent past, the raging pandemic due to COVID-19 is making the headlines, bringing about a global crisis with an inevitable spread. The use of a face mask and maintaining physical distancing is a precautionary measure as suggested by the WHO. The individuals infected with COVID-19 suffer respiratory problems accompanied by shortness of breath. The surroundings of the concerned individuals can be contaminated by their droplets carrying the virus. It is mandatory to wear a mask and follow physical distancing, yet many citizens violate the regulations. In such scenarios, frequent checks for face masks in public places and imposing fines are common. As object detection has unfolded to be an approachable biometric process, it has been widely applied in surveillance, security, autonomous driving, etc. With the rapid development of deep learning models, object detectors are highly suitable to develop social distancing and face mask detectors to administer the crowd via CCTV and surveillance cameras. The paper surveys various deep learning networks to develop such detectors. In this survey, the existing object detection models used for surveillance and people detection are analyzed. The one-stage and two-stage detectors along with their applications and performance are outlined in a comprehensive manner. Deep Learning models such as AdaBoost, Voila-Jones, variants of CNN including ResNet, VGG-16, single-shot detectors MobileNet, and versions of YOLO are discussed and compared.