Real-Time Crowd Anomaly Detection With Predictive Alerts.

The Real-Time Crowd Anomaly Detection With Predictive Alerts project demonstrates an AI-based concept for improving safety and security in crowds in a public environment. The designed system constantly analyzes surveillance videos and evaluates the crowd’s status, detecting abnormal activities, such as sudden movements, overcrowding, counterflow, panic, and other irregular behaviors. The YOLO-based person detector and crowd-density estimator provide the algorithm with insights on the number of people and their distribution, while analyzing movements and behaviors detect deviations from the norm. The proposed approach utilizes audio anomaly detection, allowing the system to receive additional information from unusual sounds, such as screams or alarms. By processing both video and audio, the crowd-monitoring system, using multimodal fusion, improves reliability and decreases fallibility in favor of a system that utilizes visual data only. Lastly, through a predictive risk engine, the algorithm assesses the rate of change of density, movement, and anomaly score, predicting the likelihood of an incident. The system is capable of classifying the context of crowds and issuing early warning alerts via a security dashboard, mobile application, SMS, alarm, or other available means. This reduces the response time of security personnel and eliminates the need for constant monitoring of CCTV feeds. Overall, this project aims for the paradigm shift of traditional surveillance systems from reactive anomaly detection to proactive crowd-safety management. By providing real-time insights, utilizing deep learning and multimodal processing, implementing an automated risk assessment engine, and issuing predictive alerts, the proposed system will offer a more reliable and faster alternative to the current safety solutions. The concept can be expanded upon with the addition of multiple camera feeds, edge computing, explainable AI, extensive data sets, and advanced predictive algorithms. The application of the designed system will be useful in railway stations, airports, stadiums, shopping malls, places of worship, smart cities, and other large public gatherings.

[1]                 L. Luo, S. Xie, H. Yin, C. Peng, and Y.-S. Ong, “Detecting and Quantifying Crowd-Level Abnormal Behaviors in Crowd   Events,” IEEE Transactions on Information Forensics and Security , vol. 19, pp. 6810-6823, 2024. [2]                 G. A. Macriga, S. Bavyatha, S. H. Abhinidhi, and N. Jagadeesh, “Crowd Management Using AI & ML,” in 2024 International Conference on Power, Energy, Control           and         Transmission          Systems (ICPECTS)2024,pp.1-5. [3]                 M. Khalid, A. R. Mohsin, J. Chandroth, J. Ali, and B.-H. Roh, “Real-Time Anomaly Detection in Crowd Surveillance Using 3D ResNet and Optical Flow,” in 2025 International Conference on Information Networking (ICOIN) , 2025, pp. 513-51. [4]                 A. Leelamony, A. Y. Ju Yang, S. Fallon, E. Fallon, R. Young, and M. E. Curran, “DeepTrackNet: Advanced Crowd Dynamics Estimation with YOLO and Deep Sort Integration,” in 2025 International Conference on Computer Science and Software Engineering (CSASE) , 2025, pp. 80-86. [5]                 M. R. Bhuiyan, J. Abdullah, N. Hashim, F. Badie, F. A. Farid, B. Duraisamy, and J. Uddin, “Towards intelligent crowd monitoring during Hajj: a novel dataset for density estimation and anomaly detection in the Tawaf area (2015-2019),” in 2025 Multimedia University Engineering Conference (MECON) , 2025. [6]                 D. Chaudhary, S. Kumar, and V. S. Dhaka, “Intelligent Anomaly Detection and Localization With Crowd Density Estimation for Improved Public Safety,” IEEE Access , vol. 13, pp. 169912-169928, 2025. [7]                 H. Hashmi, A. Kumar, and P. Singh, “Real-Time Crowd Monitoring Using Lightweight AI for Secure Zones,” 2025. [8]                 M. Gozet, M. Karakose, and A. E. Yilmaz, “YOLOv11-Based Explainable Framework For Anomaly Detection In Crowded Scenes Using Attention Fusion,” in 2025 15th International Conference on Advanced Computer Information Technologies (ACIT) , 2025, pp. 859-862. [9]                 J. Angeljulie, M. Vishal, J. Sathish, and V. Vignesh, “Real-Time Crowd Anomaly Detection in Surveillance Utilizing ML,” in Proceedings of the 10th International Conference on Communication and Electronics Systems (ICCES) , 2025, pp. 1601-1605. [10]             M. Akpulat and M. Ekinci, “Anomaly Detection in Crowd Scenes Via Cross Trajectories,” Applied Intelligence , vol. 55, article 525, 2025.