Using computer vision to measure social distancing behavior in public spaces

Project summary:

During the COVID-19 pandemic, the safe use and continued availability of public spaces depends on adherence to recommended behaviors such as physical distancing and mask wearing. However, very little data is available to monitor and assess adherence to these recommendations in a way that is simple, scalable and time sensitive. Even after the pandemic recedes, public space use is likely to be fundamentally transformed, presenting an increasing need for tools allowing policymakers to accurately and quickly measure and understand human interactions in public settings. Building on the University of Minnesota's unique personnel and computer vision resources, and leveraging existing relationships with community partners to carry out data collection, our multidisciplinary team will develop and implement a computer vision-based approach to measure physical distancing and mask-wearing behavior. Using video from multiple sources, including prospective video observation of locations in public green spaces, computer vision research datasets, and online video repositories, the research team will train deep learning models to measure key distancing metrics (average approach distance, duration of close contact, mask wearing) during encounters between two or more individuals.