YOLOv5-based weapon detection systems with data augmentation
Lucy Sumi, Shouvik Dey
Abstract
Closed-Circuit Television (CCTV) cameras in public places have become more prominent with the rising firearm-related criminal activities, such as robberies, open firing, threats at gunpoint, etc. Early detection of firearms in surveillance systems is crucial for security and safety concerns. In this paper, we present a You Only Look Once (YOLOv5)-based weapon detection system that detects different types of weapons such as rifles, pistols, knives, etc. The main objective of this work is to show the impact of data augmentation on different types of datasets and make a detailed comparative examination with the existing baseline study and other similar works in the literature. The results give new insights to consider for weapon detection systems and object detection, in general. A crisp taxonomy of the existing state-of-the-art and object detection trends over the past decades is also presented in the paper.