Implementation and Calculation of Crowd Density in Video using a Real-Time Object Detection Module

 

[ko]  Implementation and Calculation of Crowd Density in Video using a Real-Time Object Detection Module

 

Kim, Donghwan and gu, Yeri. and Kim, Han-Soo

Journal of Science Criminal Investigation 2024, vol.18, no.3,  p. 175-184

Maintaining continuous public safety in various events where large crowds gather requires quickly assessing crowd density and predicting potential casualties. The YOLO model, a 1-stage detection model capable of efficiently detecting objects and processing results in real-time is applied in this study to effectively detect and utilize crowd density observed through CCTV cameras. Utilizing various public datasets, the proposed model quickly recognizes individuals appearing on pre-installed CCTV footage and easily calculates the corresponding crowd density, with the demonstration of approximately 95% precision, 93% recall, and an F1-score of 0.89. The implemented program effectively calculates population density while also providing functionalities such as emergency alerts, graphical displays, and log storage. This study provides an efficient tool for crowd density analysis, playing a crucial role in crowd management and ensuring safety at large-scale events and public places.