AI and IoT-Based Crowd Capacity Prediction and Event Safety Management System Using Historical Event Analysis

Abstract

Large-scale gatherings in public places often face significant safety risks due to overcrowding, inadequate planning, and the lack of automated decision support. Traditional approval processes depend heavily on manual judgment, leading to frequent errors in capacity estimation and insufficient emergency preparedness. Tragic situations (e.g., Karur Temple stampede, 2024; Morbi Bridge collapse, 2022) highlight the urgent need for predictive and system-level safeguards. This paper introduces an integrated AI+IoT framework for pre-event capacity prediction and live-event risk enforcement. The AI module evaluates venue geometry, historical incidents, and event metadata to determine safe occupancy and classify risk. The IoT module combines live counts (people counters, CV- based density maps) with the predicted capacity to trigger alerts and auto-generated SOPs, including evacuation routing, steward allocation, and Public Address (PA) messaging. We describe the data format, modeling pipeline, communications stack (MQTT), evaluation metrics, and privacy measures. The hybrid framework aims to reduce overcrowding risks, minimize time-to-alert, and improve evacuation outcomes for large-scale public events.

Introduction

Large public gatherings—religious events, concerts, sports events, or political rallies—pose significant safety challenges due to high crowd densities and dynamic movements. Traditional event management often relies on manual planning, fixed occupancy limits, and human judgment, which can overlook real-world factors like spatial layout, bottlenecks, and dynamic arrivals. Past tragedies (e.g., Morbi bridge collapse 2022, Karur Temple stampede 2024, Seoul Halloween crush 2022) highlight shortcomings such as inaccurate capacity predictions, lack of live monitoring, and absence of automated emergency responses.

Proposed System:
An AI- and IoT-based Crowd Capacity Prediction and Event Safety Management System is proposed to address these gaps. It integrates Machine Learning, Computer Vision, and IoT for predictive pre-event analysis and live-event monitoring.

System Features:

1. Pre-Event Prediction and Risk Assessment:

   • Inputs: Venue floorplans, event metadata, expected attendance

   • CV module extracts spatial attributes (usable area, exits, obstacles)

   • Historical Event Database stores past records, incidents, and risk metrics

   • Capacity Estimation Model predicts safe crowd capacity CsafeC_{safe}Csafe?

   • Risk Classification Model categorizes events into Low, Medium, High risk

   • Generates a Pre-Event Safety Report for authorities

2. Live-Event Monitoring and SOP Activation:

   • IoT sensors (people counters, edge cameras, door/pressure sensors) monitor real-time crowd density

   • Cloud layer aggregates data via MQTT, calculating risk ratio R(t)=Clive(t)/CsafeR(t) = C_{live}(t)/C_{safe}R(t)=Clive?(t)/Csafe?

   • Dynamic SOP Module triggers automatic responses based on thresholds: Advisory → Alert → Emergency

   • Actions include entry control, stewards allocation, evacuation routing, PA announcements, and digital signage guidance

3. Cloud Dashboard:

   • Visualizes heatmaps, per-zone densities, inflow/outflow counts

   • Displays risk levels, SOP status, alerts, and audit logs for operators

Research Context:

• Previous work includes social-force and cellular automata models, deep learning-based crowd counting (MCNN, CSRNet), and IoT sensor networks.

• Existing systems either focus solely on surveillance or isolated safety metrics without predictive integration.

• Few solutions exist for large-scale outdoor events that combine predictive AI with real-time IoT feedback and automated operational responses.

Conclusion

The proposed AI and IoT-Based Crowd Capacity Prediction and Event Safety Management System bridges the gap between pre-event planning and real-time safety. It integrates predictive AI modeling with IoT monitoring to estimate safe occupancy, track live density via MQTT, and trigger SOPs (route optimization, steward allocation, evacuation alerts). Random Forest/XGBoost achieved low errors (RMSE = 3.2 persons/m2, MAE = 2.6 persons/m2) and CatBoost reached F1 = 0.93; the IoT layer delivered ?1.8 s alert latency. Operationally, it reduces human dependency, enforces capac- ity limits, and supports data-driven decisions. Future work: adaptive learning, UAV/drone integration, and spatiotemporal behavioral analytics for panic/flow prediction. Overall, the framework is adaptable, intelligent, and deployable, resulting in better and more secure public events.

References

[1] A. Anitha et al., “Deep Neural Architectures for Abnormal Crowd Behaviour Prediction,” Journal of Intelligent Systems, 2023. [2] Y. Chen et al., “Transformer-Based Spatiotemporal Networks for Pre- dicting Crowd Dynamics,” Neurocomputing, 2024. [3] S. Dutta and K. Roy, “People Counting and Crowd Management Using IoT and Vision Sensors,” International Journal of Smart Security Technologies, 2023. [4] S. Rahman, A. Banerjee, and P. Kumar, “An Enhanced MQTT Frame- work for Real-Time IoT Sensing in Large-Scale Smart Environments,” IEEE Internet of Things Journal, 2025. [5] D. Fernandez et al., “Next-Generation IoT Architectures for High- Density Crowd Management,” Sensors, 2025. [6] R. Fiorista et al., “CCTV Data as an Emerging Source for Rail Platform Crowding Estimation,” arXiv, 2025. [7] H. Fukuda et al., “Crowd Congestion Detection Using Edge-AI Sensors in Smart Cities,” IEEE Sensors Journal, 2024. [8] G. Gao et al., “CNN-based Density Estimation and Crowd Counting: A Survey,” Neurocomputing, 2020. [9] S. V. L. Gayathri et al., “Enhancing Crowd Management Through Behaviourally-Aware Deep Models,” Systems and Smart Applications, 2025. [10] A. Gupta and R. Verma, “Deep Learning Based Crowd Panic Detection Using Spatiotemporal Features,” Pattern Recognition Letters, 2024. [11] D. Helbing, I. Farkas, and T. Vicsek, “Simulating dynamical features of escape panic,” Nature, 2000. [12] D. Helbing, A. Johansson, and H. Z. Al-Abideen, “Dynamics of crowd disasters: An empirical study,” Phys. Rev. E, 2007. [13] M. Hernandez et al., “Adaptive Learning Models for Dynamic Crowd Risk Prediction,” Safety Science, 2024. [14] M. Hossain and A. Chakraborty, “Real-Time Crowd Density Estimation Using Drone Footage and YOLOv7,” Remote Sensing, 2023. [15] A. Zanella et al., “Internet of Things for Smart Cities,” IEEE IoT Journal, 2014. [16] P. Kamath et al., “An IoT-Cloud Integrated Framework for Emergency Evacuation and Crowd Control,” Journal of Network and Computer Applications, 2023. [17] P. Kannan, V. Jadhav, Y. Patil, K. Hyalij, and P. Bacchav, “Real-Time Crowd Monitoring and Management,” International Research Journal on Advanced Engineering Hub, 2025. [18] I. Lazarou, A. L. Kesidis, and A. Tsatsaris, “Real-Time Detection and Mapping of Crowd Panic Emergencies Using Geo-Biometrical Data and Machine Learning,” Digital, 2025. [19] Y. Li et al., “CSRNet: Dilated CNN for Understanding Highly Congested Scenes,” CVPR, 2018. [20] X. Liu et al., “A Hybrid IoT and Computer Vision Framework for Real- Time Crowd Anomaly Detection,” Expert Systems With Applications, 2024. [21] D. Mabrouk et al., “Robust Crowd Anomaly Detection Using Hybrid Ensemble Deep Models,” Scientific Reports, 2025. [22] S. S. Maharajpet and A. V. Hegde, “Intelligent Real-Time Crowd Den- sity Estimation for Proactive Event Safety,” Recent Research Reviews Journal, 2025. [23] S. Muskan et al., “Design and Implementation of a Real-Time Crowd Regulation System Using Arduino Uno and IoT Cloud Integration,” IJCRT, 2025. [24] R. Nasir et al., “An Enhanced Framework for Real-Time Dense Crowd Monitoring and Safety Management,” Applied Intelligence, 2025. [25] B. Ptak and M. Kraft, “Improving Trajectory Continuity in Drone-Based Crowd Monitoring Using Deep Discriminative Filters,” arXiv, 2025. [26] M. Rahman et al., “IoT-Based Smart Event Monitoring and Crowd Safety System,” Sensors, 2023. [27] K. Raj and S. Nambiar, “AI-Driven Urban Crowd Flow Prediction Using Multi-Modal Data,” Information Fusion, 2024. [28] M. Saleem et al., “Real-Time Crowd Behavior Tracking Using Hybrid CNN–Transformer Models,” Pattern Analysis and Applications, 2024. [29] G. K. Still, Introduction to Crowd Science. CRC Press, 2014. [30] B. D. Sunil, R. Venkatesh, and S. Todmal, “Density Estimation and Crowd Counting using Deep Learning,” arXiv, 2025. [31] J. Wang et al., “A Federated Deep Learning Approach for Privacy- Preserving Crowd Flow Prediction,” IEEE IoT Journal, 2024. [32] Y. Zhang et al., “Single-Image Crowd Counting via Multi-Column CNN,” CVPR, 2016. [33] T. Zhang et al., “Real-Time Crowd Monitoring Using Edge-IoT and Lightweight CNN Models,” IEEE Access, 2023.

Copyright

Copyright © 2025 Dr. Vairam T, Abinaya S. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.