Enhancing Safety and Security: Real-Time Weapon Detection in CCTV Footage Using YOLOv7

Abstract

In our relentless pursuit of heightened safety and security, this algorithm harnesses the formidable capabilities of the YOLOV7 deep learning model to achieve remarkable real-time weapon detection within CCTV footage.Leveraging a comprehensive dataset, the algorithm seamlessly processes CCTV frames, a pretrained YOLOV7 model, and a meticulously optimized confidence threshold.The results are striking: with an F1-score of 91 percent and a mean average precision (mAP) of 91.73 percent, it successfully identifies and annotates objects of interest.Post-processing incorporates a confidence threshold, coupled with non-maximum suppression, effectively filtering out objects with low confidence scores.Furthermore, the algorithm offers the flexibility to store frames or activate alerts based on user-defined criteria.The cycle of analysis persists for successive frames, ensuring an uninterrupted real-time vigilance.This algorithm, backed by quantifiable results, demonstrates exceptional promise for significantly enhancing safety and security across a multitude of applications.