The Application of Yolov4 and A New Pedestrian Clustering Algorithm to Implement Social Distance Monitoring during the COVID-19 Pandemic

Abstract

Based on the global spread of COVID-19 epidemic, this paper implements an application of social distance monitoring in public places, aiming to control the spread of virus by controlling the social distance of pedestrians. The application mainly uses YOLOv4 object detection algorithm and DeepSORT multiple objects tracking algorithm to detect and label pedestrians, and uses affine transformation to calibrate the scene to a more intuitive bird's-eye view. Based on the analysis of pedestrian movement, the author proposes a novel pedestrian clustering algorithm to avoid the impact of peers on the monitoring results. Finally, three indicators are selected to classify the pedestrian to analyse and evaluate the risk of virus infection in a certain place. In this paper, comparing the yolov4 algorithm with the most suitable performance of other research results, the author indirectly concludes that yolov4 is the most suitable method in the application of social distance monitoring. And the new pedestrian clustering algorithm plays a role in the application, which improves the practicability of the application. Through the experiment, it is found that the social distance in the scene can be monitored in real time and accurately, and there is a correlation between the average distance of pedestrian clusters, the density of pedestrian clusters and the infection index, which can be used to evaluate the safety of places and prevent the spread of virus.