A Fast, Automatic Risk Detector for COVID-19

Abstract

With its lethal spread to more than 200 countries, COVID-19 has brought a global crisis, affecting more than 3 crore people across the world. Viruses don’t have a cure, and this makes the population vulnerable and heavily rely on preventing the infection. Hence, following the rules of social distancing and wearing a face mask are two very essential approaches to fight against this pandemic. Motivated by this notion, this work proposes a deep learning-based framework for automating the detection of risk due to COVID19. The proposed framework utilizes YOLOv3 object detector to detect whether a person has worn a mask. In case of absence of mask, to categorize the level of risk, the person’s age category is estimated, and the result of the risk detector is displayed on the image with a bounding box. In case of multiple boxes, the framework also calculates the distance between them to check whether the rules of social distancing are being followed. The result of the YOLOv3 model is compared with popular state-of-the-art model, Faster Regionbased Convolutional Neural Network. From the experimental analysis, it is concluded that YOLOv3 object detector displays best results with respect to the trade-off between speed and accuracy.