Weapon Detection and Classification Using Deep Learning

Abstract

High gun-related crime rate poses a great threat to society in the present world. There is a serious need for systems to deal with such gun-related crimes. As CCTVs are installed in almost every part of the city, using the CCTV footage to detect the weapon is the simplest and efficient way to deal with such crimes. Unconcealed weapon detection, in images and videos, can help reduce the number of homicides due to the gun-related violence. In this work, we focus on developing a robust and automatic weapon detection system with ability to classify the detected weapon into different categories. This work provides and extensive survey on already existing weapon detection systems, weapon detection datasets, challenges in weapon detection and deep learning based object detection technologies. We have developed a new image dataset for weapon detection and classification task. The experimental analysis shows the superiority of the Faster-RCNN models over SSD models for weapon detection systems. The detection results shows how the final developed system deals with different challenges related to weapon detection and classification in real world scenarios.