Analysis of body-induced thermal signatures for social distancing monitoring

Thermal vision systems based on low-cost infrared (IR) array sensors allow to track thermal signatures induced by moving people and are promising technologies for monitoring body temperatures as well as social distancing in critical congested areas. This paper proposes a Bayesian framework for joint recognition of body temperature and location (distance and direction of arrival) of people in an indoor operational environment. Unlike conventional frame-based methods, the proposed approach exploits a statistical model for the estimation of the body distance from the IR sensors and of the direction/angle of arrival (AOA) using the body-induced thermal signatures and a mobility model for tracking multi-body motions. Compared to conventional machine learning approaches, the proposed framework processes backlogs of thermal images to prevent typical detection problems such as subject disappearance. The Bayesian method for distancing monitoring is verified experimentally with field measurements for wall mounted IR sensors.