AlertCam-An Effective Anomaly Detection System using Computer Vision

With the rising incidents of harassment and violence in India, particularly against women, this paper presents a real-time violence detection system using YOLO image detection models (YOLOv7, YOLOv8, and YOLOv9). The models were trained on a carefully curated and manually annotated dataset using Roboflow, ensuring robust detection across diverse scenarios. Transfer learning and hyperparameter tuning were applied to improve accuracy. The system was evaluated using key performance metrics, including mean average precision (mAP), Intersection over Union (IoU), precision, and recall, achieving 80% precision and 50–60% recall. Extensive manual testing on real-world surveillance footage and additional video sources further validated its effectiveness. Results indicate that YOLOv8 outperforms other versions in balancing speed and accuracy. This system has strong potential for deployment in law enforcement, public surveillance, and safety-critical applications.