Deep learning based crowd monitoring and person identification system

Abstract

With the recent outbreak of COVID-19 and other pandemics that might occur in the future, we have realized how crucial it is to monitor crowd behavior and social distancing in public places. This paper introduces a deep learning-based system for crowd monitoring and person identification, addressing the challenges posed by similar kinds of pandemics. The approach proposes a combination of person tracking and distance measurement systems which can be implemented in various public places. In this paper, we come up with two approaches, the first one is based on utilizing Faster R-CNN for person identification along with The Simple Online and Real-time Tracking (SORT) algorithm and the other approach is YOLOv8 for person detection, coupled with the DeepSort-based tracking algorithm which also includes a logging system. Moreover, to implement the system, a custom dataset is created for evaluation and to tackle perspective correction and person-only detection issues. After using our custom dataset for evaluation and also evaluating the performance and robustness, we propose our DeepSort-based second approach as the better system. Moreover, our proposed system achieved over 90% accuracy in human detection and distance estimation.