Real-Time Detection of Abnormal Behavior Using YOLO

Crowd management and safety are critical concerns during large-scale events like Hajj, where millions of people gather in a confined space. The complexity of monitoring such massive crowds and detecting abnormal behaviors in real-time poses a significant challenge. This project aims to enhance crowd safety at Hajj by employing a deep learning-based real-time detection system using YOLOv12 for identifying and classifying abnormal behaviors in crowded environments. The system leverages YOLOv12's advanced object detection capabilities, classifying various behaviors such as Non_Pedestrian, Opp_Direction, Running, Sitting, Sleeping, and Standing, as well as the Diff_Direction class to detect abnormal movement patterns that may indicate potential safety risks.

The project utilizes a dataset with diverse behavioral classes and employs YOLOv12 to analyze visual data in real-time, providing immediate alerts for suspicious activities. The results show high precision and recall rates across different classes, with the highest performance in detecting Sleeping (precision: 0.981, recall: 1) and Non_Pedestrian (precision: 0.892, recall: 0.822). The system aims to provide a scalable solution for crowd safety, ensuring effective monitoring and timely intervention in case of abnormal behavior, contributing to the safety and security of pilgrims during Hajj. By combining real-time deep learning-based detection with advanced object detection techniques, this project highlights the potential of AI-powered systems in enhancing crowd safety management.

Keywords: Crowd Safety, YOLOv12, Abnormal Behavior Detection, Deep Learning, Object Detection, Real-Time Monitoring, Hajj, Anomaly Detection, Safety Management, Large-Scale Events, Machine Learning, Computer Vision.