AI-Enabled Crowd Monitoring and Emergency Public Safety Alert System Using YOLOv8 and LSTM

Abstract

Rapid urbanization and increasing public gatherings have created major challenges in crowd management and public safety. Traditional surveillance systems depend heavily on manual monitoring, which may lead to delayed emergency response during overcrowding, panic situations, or stampedes. This paper proposes an AI-enabled crowd monitoring and emergency public safety alert system using deep learning techniques for real-time crowd analysis and disaster prevention. The proposed system employs the YOLOv8 algorithm for accurate human detection and crowd density estimation, while Long Short-Term Memory (LSTM) networks are used for behaviour analysis and abnormal activity detection. The system continuously monitors crowd movement patterns and automatically generates alerts when overcrowding or unusual behaviour is detected. Experimental evaluation was conducted using crowd surveillance video datasets containing approximately 5,000 annotated frames collected from public environments. The proposed model achieved 95.8% detection accuracy with real-time processing speed of 28 FPS, demonstrating effective performance in dense crowd environments. The system provides automated monitoring, rapid emergency alerts, and reduced response time, making it suitable for railway stations, stadiums, airports, political rallies, and smart city surveillance applications.