Anti-Stampede Detection and Prevention Using Deep Learning

Stampedes represent a significant risk in high density areas, often resulting from a failure to detect early warning signs of crowd turbulence. This paper presents an automated Anti-Stampede System designed to bridge the gap between surveillance and emergency response. Leveraging the InceptionV3 Deep Learning architecture within a distributed edge computing framework, the system analyzes live CCTV feeds to identify stampede inducing anomalies. By shifting computational heavy lifting to edge nodes, we achieve low-latency real-time analysis while reducing the central network load. The system integrates a multi-channel alerting mechanism that utilizes WhatsApp notifications, localized audio alarms, and precise Positioning data. Experiments confirm that the InceptionV3 model reaches 94.6% accuracy, exceeding VGG16-based approaches and offering a dependable solution for public safety.