Smart Camera for Enforcing Social Distancing

Abstract

The COVID-19 pandemic presents an unprecedented challenge to public health, food systems and the demand and supply chains. “Coronavirus” spreads when an infected person coughs, sneezes or talks, and droplets from their mouth are launched into the air and inhaled by people in the vicinity. Mid-2021 witnessed the production and supply of effective vaccines against Coronavirus, and around 4.5 billion vaccine doses have been utilised globally, reducing fatalities significantly. Given the Government’s plans to ease quarantine restrictions for schools, offices, and public places, Social Distancing has become even more critical than ever before. This project incorporates Computer Vision techniques using the high-performance YOLOv4 library, DSFD Face detector, Deep Learning Darknet and Pre-trained ResNet models, and RaspberryPi to create a plug-and-play extension for CCTV cameras established in public places. The system uses the frame by frame information of CCTVs to detect people and classify violations of Social Distancing norms. The device also performs real-time Face Mask Detection, and this technique is robust to varying geometries of face masks and degrees of natural illumination. In case of a detected violation of Social Distancing norms, a buzzer blares in the background. The timestamp of violation with the snapshot of the frame highlighting the associated people is sent to a database and emailed to a centralised server for further investigation.