Multi-Modal Real Time Surveillance System Architecture for Intelligent Crowd Control Using YOLOv8

Abstract

High-density gatherings such as the Kumbh Mela pose severe crowd-safety challenges, motivating advanced real-time surveillance solutions. This paper presents a novel multi-modal crowd-control system that integrates video-based detection using YOLOv8, embedded pressure sensors, and thermal imaging to provide robust crowd metrics through sophisticated sensor fusion techniques. Our architecture employs spatial-temporal alignment of heterogeneous sensor streams, followed by confidence-weighted Kalman filtering (60% visual, 30% pressure, 10% thermal) to generate reliable crowd state estimates. The system incorporates a comprehensive analytics engine performing density estimation, flow analysis, bottleneck detection, route optimization, and predictive forecasting. Experimental validation demonstrates superior performance over single-modality approaches, achieving 94.7% detection accuracy and sub-100ms response times. The framework addresses critical limitations in conventional surveillance systems while providing actionable insights for proactive crowd management in ultra-large-scale events.