Weapon Detection using Mask R-CNN and ResNeXt

We are living in an era marked by security challenges. Traditional surveillance systems are struggling to keep up with new ways of doing crime. The development of advanced weapon detection system has become important in fight against crime. In this study, a weapon detector system using the Mask Region Convolutional Neural Network (Mask R-CNN) algorithm with ResNeXt as a backbone architecture is proposed. We replace the default Mask R-CNN backbone with ResNext to improve the quality of the features extracted because of ResNext's good cardinality and grouped convolutions. The model generates a high quality segmentation mask for each object instance in the image. The system uses the COCO dataset to get pictures of weapons. We evaluated our system using various performance metrics, our experimental results shows that combination of R-CNN and ResNeXt is one of the most effective ways to detect weapons.