Real-Time Stampede Monitoring and Detection System using Deep Learning and Computer Vision

T. Likhitha
Student, Dept of CSM Sri Venkateswara College of Engineering Tirupati
likhitha@svce.edu.in
P. M. Kishore
P. Sai Charan Student, Dept of CSM Sri Venkateswara College of Engineering Tirupati
saicharan@svce.edu.in
B. Harsha Vardhan
Student, Dept of CSMSri Venkateswara College ofEngineeringTirupati
harshavardhan@svce.edu.in
N. Phani Kumar
Student, Dept of CSMSri Venkateswara College ofEngineeringTirupati
kishore@svce.edu.in
Assistant Professor, Dept of CSMSri Venkateswara College ofEngineeringTirupati
phanikumar@svce.edu.in

Abstract—Crowd management in densely populated environments is a critical public safety concern due to the increasing frequency of stampede incidents resulting in severe casualties. This paper presents a real-time Stampede Monitoring and Detection System that leverages state-of-the-art computer vision and deep learning techniques to analyze crowd behavior and detect potential stampede situations before they escalate into disasters. The proposed system utilizes YOLOv8-Pose for precise human detection and skeletal keypoint extraction, enabling posture analysis for identifying falls and crouching behaviors indicative of crowd panic. Speed tracking, trajectory analysis, and directional flow evaluation are integrated alongside a composite risk engine that computes a unified risk score ranging from 0–100%, classifying situations as Safe, Warning, or Critical. Video input from surveillance cameras is processed in real time to detect anomalies including sudden crowd surges, unusual motion patterns, and high-density clustering. Upon detecting a potential risk threshold breach, the system generates immediate alerts to authorities via a Streamlit-based interactive web dashboard for timely intervention. Experimental results demonstrate improved accuracy in identifying critical crowd scenarios with significantly reduced false positives. The systemis scalable and deployable across public venues including stadiums, railway stations, religious gatherings, and festival grounds, substantially enhancing public safety and disaster prevention capabilities. Index Terms—Crowd Monitoring, Stampede Detection, Deep Learning, Computer Vision, YOLOv8-Pose, CNN, Anomaly Detection, Optical Flow, Risk Engine, Smart City Surveillance, Public Safety.