Physical Distance and Crowd Monitoring System using YOLOv3

Abstract

The World Health Organization advised the public that physical distancing is one of the health protocols that can minimize the spread of coronavirus disease 2019 (COVID-19). The protocol requires people to adhere to a one-meter distance from each other in public areas, thus avoiding the possible crowd formation and further spread of the virus. A software was developed to monitor the physical distance and the crowd density of a specified area or the region of interest using You Only Look Once Version 3 (YOLOv3). Video recordings, captured using mobile phones, were extracted into frames. Each video frame is then processed to a YOLOv3 model for further object detection (here-human) and implementation of physical distancing monitoring. The selected area's crowd density is also computed while considering physical distancing guidelines. If the violations in physical distance or crowd density become alarming, an email will be sent to the authorities alerting them about the occurrence of health protocol violations. Based on careful evaluation, physical distancing and crowd density violation detection has an average of 0.86 for precision, 0.81 for recall, 0.83 for F1-score, and 0.83 for accuracy. The software also successfully alerted authorities via email of the exceeding violations. The efficiency and simplicity of this approach present possible solutions for the current pandemic situation.