Social Distance Approximation on Public Transport Using Stereo Depth Camera and Passenger Pose Estimation

Abstract

In order to effectively balance enforced guidance/regulation during a pandemic and limit infection transmission, with the necessity for public transportation services to remain safe and operational, it is imperative to understand and monitor environmental conditions and typical behavioural patterns within such spaces. Social distancing ability on public transport as well as the use of advanced computer vision techniques to accurately measure this are explored in this paper. A low-cost depth-sensing system is deployed on a public bus as a means to approximate social distancing measures and study passenger habits in relation to social distancing. The results indicate that social distancing on this form of public transport is unlikely for an individual beyond a 28% occupancy threshold, with an 89% chance of being within 1–2 m from at least one other passenger and a 57% chance of being within less than one metre from another passenger at any one point in time. Passenger preference for seating is also analysed, which clearly demonstrates that for typical passengers, ease of access and comfort, as well as seats having a view, are preferred over maximising social-distancing measures. With a highly detailed and comprehensive set of acquired data and accurate measurement capability, the employed equipment and processing methodology also prove to be a robust approach for the application.

Keywords

• social distancing
• computer vision
• stereo camera
• pose estimation
• RaspberryPi
• public transport