Smart Al-Driven Crowd Density Monitoring and Stampede Prevention System

DOI

10.2991/978-94-6239-693-7_54How to use a DOI?

Keywords

crowd monitoring; deep learning; density estimation; YOLOv8; LSTM; real-time analytics; surveillance; prevention; forecasting

Abstract

Manual crowd monitoring at public gatherings, religious events, and transportation hubs often causes delays, inaccurate estimates of crowd density, and limited awareness of the situation, especially in busy environments where crowd behaviour changes quickly. This paper introduces a smart vision-based system that automatically estimates crowd density and gives early warnings to prevent stampedes using deep learning and real-time analytics. The system uses the YOLOv8-S model for person detection because it balances speed and precision well. To train the model, we used two different datasets: the public Shanghai Tech crowd dataset and our custom Crowd Safe dataset, which reflects various crowd densities, lighting conditions, and camera angles. Preprocessing modifies input frames to 640×640 pixels, while techniques like brightness adjustment, blurring, and perspective changes help the model work well across different scenarios. Detected individuals are analysed through a spatial clustering and heatmap-based density estimation module, which classifies crowd levels as low, moderate, or high. A risk assessment engine then combines temporal data with LSTM-based trend analysis to predict where overcrowding might occur. The system sends alert notifications to authorities and automatically activates control mechanisms like gate regulation or voice announcements. Experimental results show that the system accurately detects crowds and estimates density with more than 93% accuracy in different environmental conditions, making it suitable for live surveillance feeds. Current limitations include reduced performance in very low-light situations and cases of partial occlusion. Future research will aim to include thermal imaging and sensors to enhance robustness and improve early anomaly detection. This study shows that combining deep learning detection and temporal forecasting offers a practical and efficient way to manage crowds and prevent stampedes.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Madena Venu Gopala Rao
AU  - L. Kartheesan
AU  - Tholichuri Ravi Teja
PY  - 2026
DA  - 2026/06/16
TI  - Smart Al-Driven Crowd Density Monitoring and Stampede Prevention System
BT  - Proceedings of the International Conference on Intelligent Systems for a Sustainable Future (ISSF 2026)
PB  - Atlantis Press
SP  - 547
EP  - 557
SN  - 2589-4919
UR  - https://doi.org/10.2991/978-94-6239-693-7_54
DO  - 10.2991/978-94-6239-693-7_54
ID  - Rao2026
ER  -