Proposed Methodology for Real-Time Pistol Detection System

Handheld weapon violence has increased global crime rates. Numerous incidents show that handguns are commonly used as the primary weapon in break-ins, robberies, etc. Such crimes can be reduced by detecting and monitoring suspicious activity so that law enforcement can act quickly. CCTV cameras are critical security measures that aid in the resolution of this issue. CCTV is installed in every public place for safety, crime investigation, and other detection-based security. A manual surveillance system demands the use of a human eye to detect malicious activity, notifies security personnel steadily, and is susceptible to human error. Pistols can be detected using the You Only Look Once (YOLO) Object Detection algorithm. This paper provides a thorough introduction, network architecture, and internal mechanisms of the three versions of the YOLO algorithm, namely YOLO3, YOLOv4, and YOLOv5. In addition, a comparative evaluation of the performance of these models was carried out by analyzing the implementation outputs and researching past research works. Precision and recall are more crucial than accuracy when it comes to object detection, hence the models were assessed in terms of them. Yolov5 outperformed the other two versions, having a higher F1-score and mean average precision. We go over some of the challenges encountered in pistol detection, as well as some potential future opportunities and new perspectives on how to address them and improve the system.