Intelligent Crowd Monitoring and Early Warning Framework for Stampede Prevention Using YOLOv8m and CNN-LSTM Architecture

Abstract

Railway stations, religious events, and other large public gatherings are prone to overcrowding, sometimes resulting in stampede situations. The existing systems for monitoring crowds primarily rely on manual monitoring, which prevents them from producing early warnings for potentially dangerous situations. In this paper, the authors describe a machine vision-enabled crowd monitoring and early warning framework that is based on YOLOv8m, CNN, and LSTM models with the ability to perform real-time analysis of crowds for the purpose of predicting the risk of a stampede. The framework uses real-time video streams from surveillance cameras for the processes of crowd detection, density estimation, motion analysis, flow direction tracking, and prediction of temporal behaviour. Results demonstrate that the proposed framework successfully identifies abnormal crowd behaviour and predicts the potential for extreme or dangerous situations prior to their occurring. The proposed system enhances overall crowd safety within a high-density area, enhances the accuracy of detecting abnormal behaviour, and improves the ability to provide early warning capability of a potentially dangerous crowd situation.