AI-Powered Threat at Weapon Detection Using Surveillance Cameras

Abstract

This project proposes an AI-powered threat detection system capable of automatically detecting weapons in real time from CCTV footage, specifically focusing on pistols. Security in modern society is a growing concern, especially for countries aiming to create a safe environment for investors and tourists. While Closed Circuit Television (CCTV) cameras are widely used for surveillance, they still depend on human oversight. This project addresses the need for an automated system capable of detecting illegal activities, specifically weapons, in real-time using CCTV footage. Current deep learning techniques, despite advancements in hardware and software, face challenges such as occlusions, viewing angles, and varied environments. This project proposes a solution utilizing state-of-the-art deep learning algorithms for weapon detection, focusing on pistol detection using a custom dataset created from various sources, including manual collections, YouTube, GitHub, and public databases. Through testing multiple models, YOLOv4 emerged as the most effective with an F1-score of 91% and mean average precision (mAP) of 91.73%. Keywords--Weapon Detection, Deep Learning, Object Detection, Artificial Intelligence, Computer Vision, Real-time Surveillance