Real time Weapon Detection Using Convolutional Neural Network

Abstract

Security is a big issue these days. People are becoming violent and the tendency to be armed is being noticed. Carrying a pistol or knife is now very common. Security guards are trying their best, but we think technology can be used very effectively in this case, such as machine learning. We designed this model to detect real-time weapons. It takes video input from a camera, CCTV, or any other device and can detect if there are any weapons. Here we use Deep Learning techniques based on CNN (Convolutional Neural Network). Since so many algorithms exist, it was difficult for us to choose an algorithm that suited our work. Finally, we choose “YOLOv3”, an algorithm that uses convolutional neural networks for object detection. Then we faced the biggest hurdle, "DATA". There was not enough data on the internet to train our model. Then, we decide to produce the data ourselves. We used the VLC media player to collect images from videos and then used LabelImg, a python library to label those images. After meeting all the pre-requirements, we started writing our scripts and training the model. Eventually, we were able to build what we expected. We have built a model that can detect weapons in real-time and it is very much possible to send signals to specific destinations or play an alarm.