Beyond Counting: A Framework for Real-Time Affective and Density-Based Crowd Intelligence

Traditional crowd surveillance systems often overlook the collective emotional state of a crowd, a crucial indicator for anticipating safety risks like panic or unrest. This paper presents a real-time framework that integrates crowd density monitoring with affective state analysis for deeper situational awareness. The system leverages deep learning models for facial detection, tracking, and emotion classification, powering a dual-mode anomaly detection engine that simultaneously identifies overcrowding and spatial clusters of negative emotions. Operating at 10 FPS, the framework achieves an overall emotion recognition accuracy of 85.2% and high reliability in detecting density anomalies (0.98 precision). This work bridges a critical gap between physical and affective crowd intelligence, offering a robust solution for next-generation surveillance systems.