Framework for timely perception of spatiotemporal crowd congestion risk in public spaces based on video pedestrian tracking and geographic mapping
Shaojun Liu, Ling Zhang, Junlian Ge, Weitao Li, Yi Long · GIScience & Remote Sensing · 2025
As urbanization intensifies, congestion and safety concerns in various public venues have garnered significant societal attention, presenting substantial challenges for public safety management and emergency response. To address this issue, developing a technological framework capable of automatic perception, analysis, and early warning of crowd dynamics is urgently needed. Current technologies face hindrances such as low data precision, limited generalization across diverse scenarios, incomplete crowd coverage, and sluggish responsiveness. The popularity of sensor networks and the rapid development of computer vision technology have made it possible to collect and perceive spatial and temporal information ubiquitously, creating valuable opportunities for the rapid analysis of geographic phenomena and timely detection of potential problems. This study drew on the concepts of social comfort distance and spatial carrying capacity by employing high-coverage surveillance video streams as the data source. We introduced a framework for crowd activity perception and scene-adaptive congestion risk assessment based on a multi-object tracking model. This framework established a mutual mapping between the image and geographic spaces, facilitating precise spatiotemporal identification and grading of congestion. The validity of the method was demonstrated across various scenes in a tourist area through a comprehensive analysis of its perception accuracy and efficiency. The flexible and modular architecture of the proposed technology paves the way for its broader application and offers an effective solution for refined crowd management in public spaces.