Key points are not available for this paper at this time.
Large-scale public gatherings in India, such as temple festivals, railway stations, and political rallies, often result in dangerously dense crowds, posing serious risks to public safety. Traditional surveillance methods, which rely on manual CCTV monitoring, are error-prone and insufficient for real-time crowd control. This paper presents Smart Crowd, an AI-driven, multi-sensor framework for real-time crowd density monitoring and predictive management tailored to Indian public hotspots. The proposed system combines overhead CCTV feeds and drone footage with non-visual inputs such as Wi-Fi / Bluetooth probe signals to enhance crowd estimation accuracy. A deep learning pipeline integrates YOLOv5 for object detection, CSRNet-style CNNs for density mapping, and LSTM models for congestion prediction. The system operates entirely on edge devices (e.g., Jetson Xavier NX), achieving 20+ FPS with <45 ms latency, ideal for rural or bandwidth-constrained environments. Experimental results demonstrate improved performance over baseline models, with a 28% reduction in MAE and timely alerts issued up to 5 minutes before critical congestion events. Smart Crowd offers a scalable, real-time solution for public safety, enabling authorities to respond proactively and reduce crowd-related risks.
B et al. (Wed,) studied this question.