Real-Time Crowd Monitoring Using Lightweight AI for Secure Zones

Proactive security in red-alert areas, airports, and critical infrastructure requires intelligent real-time monitoring systems with high-precision and low-latency crowd-related anomaly detection. In this paper, a light-weight AI-driven crowd monitoring framework is introduced that exploits hybrid architecture by incorporating a Convolutional Neural Network (CNN) for spatial feature extraction and a Support Vector Machine (SVM) classifier for temporal anomaly detection. The model is optimized for edge deployment, prioritizing lower computational overhead and power consumption, to fit the surveillance hardware into embedded devices. The system is designed to analyze video feeds in real-time to identify anomalous behaviors like rapid crowd gathering, intrusion, and unusual motion patterns. Experimental results on standard datasets such as UCSD Anomaly and ShanghaiTech demonstrate that the system suggested in this work achieves very high detection accuracy (over 92%) while maintaining real time inference speed. The results highlight the scalability of the system for use in security-critical systems with limited computational capacity. Comparisons with state-of-the-art approaches such as YOLOv5, OpenPose, and density estimation methods validate the hybrid model's efficacy and efficiency.