AI-Based Real-Time Crowd Monitoring and Alert System Using YOLOv8 and DeepSORT

Abstract

Surveillance of crowd action in high-density areas is critical to maintaining safety and preventing dangerous conditions due to overcrowding. The objective of this study is to devise an intelligent system that is capable of detecting, tracking, and assessing risks of a crowd with high reliability in real time. The innovation of the work is that the advanced object detection, multi-object tracking, crowd density estimation, and automated alert mechanisms are built into a single edge-based framework. The proposed CROWD SENSE NEO model uses YOLOv8 to perform precise human detection and DeepSort to track the identity of the person across frames. It also integrates density-based risk classification to group crowd conditions into various levels of safety and provide automatic notifications in case of critical situations. The system is effective in processing the live video streams and also it offers the continuous monitoring through an interactive dashboard with minimal latency. Experimental findings also reveal that the proposed framework can achieve an accuracy of 97.4% in terms of accuracy and can be used to complete the crowd monitoring tasks with high accuracy. The framework has proven to be a stable and scalable solution to the application of the situational awareness, reduction of response time, and support of informed decision-making.