Enhancing Crowd Safety in Mass Gatherings Through Real-Time Monitoring Using Raspberry Pi

Abstract

Crowd safety in mass gatherings remains a critical challenge for event organizers and public safety officials, with manual monitoring proving insufficient and cloud-based computer vision systems incurring prohibitive infrastructure costs. This paper introduces an edge-AI framework for real-time crowd density monitoring deployed on Raspberry Pi 4 hardware, enabling low-cost, distributed surveillance without server dependency. The proposed system integrates the Histogram of Oriented Gradients (HOG) detector with adaptive thresholding for crowd density estimation, achieving 87.3% detection accuracy at 15 FPS while consuming only 410 MB RAM. Experimental deployment in a real-world venue demonstrates alert propagation latency of 230 ms, making the system suitable for decentralized crowd monitoring in resource-constrained environments. Our contributions include: (1) an optimized edge computing pipeline for commodity hardware, (2) a threshold-adaptive risk classification algorithm, and (3) validation of real-time performance on embedded plat- forms. Results show 3.2× lower power consumption compared to cloud-dependent systems and 67% reduced infrastructure costs. The framework is deployable at scale across multiple venues with minimal IT overhead, advancing practical solutions for crowd safety in developing nations and resource-limited settings.