Firearm Detection Using YOLOv8 Deep Learning Model

Abstract

Firearm detection in surveillance imagery is an important computer vision task for automated security monitoring. This study presents a deep learning-based firearm detection system using the YOLOv8 (You Only Look Once, version 8) object detection framework. A dataset of 1,600 firearm images was prepared and annotated to train and evaluate the detection model under varying visual conditions, including changes in illumination, viewing angle, and partial occlusion. The proposed system performs real-time firearm detection from images and video streams and identifies the location of detected firearms using bounding boxes. Post-processing is employed to improve the interpretation of detection results and support alert generation. The model is designed with real-time deployment in mind and can be adapted for surveillance applications where computational efficiency is required. Experimental evaluation on the prepared dataset achieved an mAP@0.50 of 99% indicating effective firearm recognition under the evaluated conditions. The results demonstrate the potential of YOLOv8-based object detection for automated firearm detection in surveillance environments.