Comparison of Physical Distances Between Pedestrians on a Street in the Central Area of Osaka City Before and After the Covid-19 Pandemic Based on Deep Learning Techniques

Atsushi Takizawa, Haruka Narumoto, Shinpei Ito, Nagahiro Yoshida

Keywords: COVID-19, Physical distancing, Pedestrian detection, Osaka, Yolo, DeepSort, Average nearest neighbor, Ripley's K-function

TL;DR: Comparison of Physical Distances Between Pedestrians on Streets Before and After the Covid-19 Pandemic

Abstract: COVID-19 has been spreading worldwide since 2020. Although the World Health Organization recommends maintaining a physical distance of at least 1 m among people, the Japanese government recommends 2 m. In this study, we used deep learning and other techniques to statistically compare and verify the change in physical distance between pedestrians on a sidewalk in a large Japanese city before and after the COVID-19 pandemic. A video-based approach was used to accomplish this. The video before the COVID-19 pandemic was recorded in October 2018 in the Namba area of Midosuji, Osaka City. For comparison, new videos were recorded at the same location in October 2020. YOLOv3 SPP was applied to automatically extract a large number of pedestrians on the street. Three observation areas were set on the sidewalk within the target area, and the physical distances between the pedestrians were measured. Two indices were used to measure the physical distance: the average nearest neighbor and Ripley’s K-function. Thus, the change in the physical distance between people on the street, before and after the COVID-19 pandemic, could be quantitatively and statistically compared. The results showed an increase in physical distance after the COVID-19 pandemic, which depended on the state of behavior, density, and human relations.