Copyright © 2026 Authors retain the copyright of this article. This article 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.

@article{209537,
        author = {Krunal Vighre and Ms. Radhika Mokati and Mahesh Khadilkar and Gokul Tong and Yogesh Yadav and Ritik Dhodharmal},
        title = {Real-Time Stampede Risk Prediction and Alert-Driven Officer Dispatch Using YOLO-Based Crowd Density Analysis},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {5},
        pages = {2138-2147},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=209537},
        abstract = {Crowd crushes at festivals, pilgrimages, railway stations and stadiums are usually recognised only after people are already hurt, and even a correct early warning is of little value if it never reaches someone able to act. This paper describes a software system that joins detection and response in a single pipeline. A YOLO-based person detector processes uploaded recordings or live camera feeds; detections are binned into a 6×8 spatial grid, and a rule-based analyser combines the peak head-count in any one cell with the frame-to-frame growth of the total count to assign one of five risk levels. A decaying heat accumulator gives operators a visual overlay. When the level reaches HIGH or CRITICAL, the alert is stored, authorities are notified by e-mail, and a dispatch engine offers the incident to the nearest on-duty field officer for a limited time, moving to the next-nearest officer after a decline or timeout. The officer’s phone application reports position, receives Web Push notifications, and advances the incident through accepted, en-route, arrived and resolved states. We present the architecture, the thresholds and the state machine, report 104 scripted consistency checks on the dispatch logic (all passing), and report a preliminary evaluation on 12 labelled video windows, in which the system’s alerts agreed with human labels in 66.7% of windows (recall 0.400 for RISK and 0.857 for NO_RISK). The risk analyser is not yet learned from data and the evaluation set is very small, so we state the resulting limitations plainly.},
        keywords = {crowd safety, stampede prediction, person detection, YOLO, crowd density, risk assessment, emergency dispatch, Web Push, real-time video analytics},
        month = {October},
        }