Physical Distancing Detection using YOLO v3 and Bird's Eye View Transform

Abstract

One regulation that has been established by governments in most countries to curb the spread of Covid-19 is physical distancing. However, many people still ignore the importance of this regulation. Thus, it is important to develop a system that can help enforcing this regulation. In this paper, we propose a system that can automatically detect the presence of humans in a video frame and measure their distances from each other. Object detection is performed using YOLO v3 and the accuracy of distance measurement is enhanced using Bird's Eye View Transformation. Our experiments show that using this transformation yields an accuracy improvement of up to 20.93% compared to the performance of the system without transformation (i.e., from 74.42% to 95, 35% accuracy).