Crowd Density Mapping and Anomaly Detection using YOLOv8 and DEEPSORT

This study proposes an effective real-time crowd monitoring framework that integrates YOLOv8 for object detection and DeepSORT for multi-object tracking and behavioral analysis. The system is capable of identifying critical anomalies—such as panic, physical altercations, and prolonged inactivity—while also categorizing crowd density into low, medium, and high levels. The visual output, presented through a Google Maps-based interface, enhances usability and situational awareness for both authorities and the general public. With GPU acceleration, the framework achieves a real-time performance of 10–20 frames per second, ensuring responsive and smooth monitoring suitable for live surveillance applications. Despite demonstrating a precision rate of approximately 86%, the system has certain limitations, particularly under conditions involving heavy occlusion or extremely dense crowds where detection accuracy may decline. Additionally, effective anomaly recognition is dependent on a minimum number of individuals within the camera frame to accurately interpret motion and spatial cues. The current approach focuses solely on person-based behaviors, overlooking context- or object-specific anomalies. Future work could address these challenges by incorporating spatio-temporal attention mechanisms, transformer-based models, and multi-camera coordination to enhance robustness, scalability, and contextual awareness in complex environments.

[1]   S. Author and S. Roy, “A comprehensive survey on crowd anomaly detection using deep learning,” 2024. [Online]. Available: https://doi.org/10.13140/RG.2.2.24390.48961 [2]   S. Shukla, B. Kumar, and H. Tiwari, “A comprehensive survey on abnormal crowd behaviour,” Northern Economic Review, vol. 15, no. 1, 2024. [Online]. Available: https://nerj.org/. DOI: https://doie.org/10.0130/Nerj.2024509272 [3]  D. Tank, S. G. Patel, and D. S. Pandya, “Recent advancements in deep learning for crowd anomaly detection: A comprehensive survey,” Int. J. Intell. Syst. Appl. Eng., vol. 12, no. 21s, pp. 2889–2902, 2024. [Online]. Available: http://www.ijisae.org [4]  M. H. Sharif, L. Jiao, and C. W. Omlin, “Deep crowd anomaly detection: State-of-the-art, challenges, and future research directions,” Artif. Intell. Rev., vol. 58, p. 139, 2025. DOI: 10.1007/s10462-024-11092-8 [5]  H. Gaikwad, S. Jadhav, N. Gunjal, S. Survase, and K. Shinde, “Real-time crowd monitoring system,” Int. Res. J. Mod. Eng. Technol. Sci., Maharashtra, India, 2024. DOI: https://www.doi.org/10.56726/IRJMETS38950 [6]  M. H. K. Khel, K. A. Kadir, S. Khan, M. Noor, H. Nasir, N. Waqas, and A. Khan, “Realtime crowd monitoring—estimating count, speed and direction of people using hybridized YOLOv4,” IEEE Access, 2024. [7]  S. Altowairqi, S. Luo, and P. Greer, “A review of the recent progress on crowd anomaly detection,” Int. J. Adv. Comput. Sci. Appl., vol. 14, no. 4, 2023. DOI: 10.14569/IJACSA.2023.0140472 [8]  G. Gao, J. Gao, Q. Liu, Q. Wang, and Y. Wang, “A survey of deep learning methods for density estimation and crowd counting,” Vicinagearth, vol. 2, no. 2, 2025. DOI: 10.1007/s44336-024-00011-8 [9]  L. Deng, Q. Zhou, S. Wang, J. M. Górriz, and Y. Zhang, “Deep learning in crowd counting: A survey,” CAAI Trans. Intell. Technol., 2023. DOI: 10.1049/cit2.12241 [10]   M. J. Babar, M. Husnain, M. M. S. Missen, et al., “Crowd counting and density estimation using deep network—A comprehensive survey,” TechRxiv, Jun. 2, 2023. DOI: 10.36227/techrxiv.23256587.v1