An Efficient Deep Learning-Based Firearm Detection System Using Yolov8

Abstract

Firearm detection is critical in ensuring public safety and security, particularly in sensitive areas such as schools, airports, and public gatherings. Group 1 represents the YOLOv4-based firearm detection system, tested across 26 models with varying detection thresholds and accuracy levels. Key parameters like accuracy, latency, and detection time ensure effective firearm detection under diverse scenarios. Group 2 focuses on the proposed YOLOv8 model, which utilizes a public dataset to improve real-time detection accuracy and efficiency. YOLOv8 performs better than YOLOv4 in terms of detection speed, accuracy, and dependability. YOLOv8 outperforms YOLOv4 in terms of accuracy (77.80–88.50%), error rate (0.65–0.73), and detection times (1.50–2.20 seconds), while achieving higher accuracy (94.80–98.00%) and a lower error rate (0.34–0.41).achieving a significance value of 0.005.  This work contributes to the growing need for effective firearm detection tools and highlights YOLOv8's potential to enhance security solutions with improved object discrimination capabilities