ADRAS: Airborne Disease Risk Assessment System for Closed Environments

Airborne diseases are easy to spread in any population. The advent of COVID-19 showed us that we are not prepared to control them. The pandemic has drastically posed challenges to the daily functioning of public and private establishments. In general, while there have been several approaches to reduce the potential risk of spreading the virus, many of them rely on the commitment that people make, which - unfortunately - cannot be constant, for example, wearing a facemask in closed environment at all times or social distancing. In this work, we propose a computer vision system to determine the risk of airborne disease spread in closed environments. We modify and implement the Wells-Riley epidemiological equation. We also evaluate and implement models for facemask and person detection from OpenVino. For mask detection, we applied transfer learning and obtained the best performance for a model based on MobileNetV2. The generated data from several devices is visible in a web platform to monitor multiple areas and locations. Finally, an OAK-D camera and a Jetson device are embedded in a end device meant to monitor a closed environment and send spread risk data continually to the web platform. Our results are promising as we obtained up to 88% of accuracy for the person detection task and up to 57% of mAP for the facemask task. We expect this paper to be beneficial for developing new control measurements and prevention tools to prevent airborne contagion.