Real-time Social Distancing Detection System with Auto-calibration using Pose Information

In this demo, we present a real-time social distancing system with auto-calibration using human pose information. Our system first calculates geometric parameters of a camera in 3D space, i.e. position and rotation, then, measures absolute distances between pedestrians. Detection results are visualized as 3D circles in input images and a bird’s eye view. All processing steps are completely automatic, therefore, our system can be applied for uncalibrated surveillance cameras, which are already installed in town. A demo video is available in the supplemental material.