Automatic Gun Detection From Images Using Faster R-CNN
Rana M. Alaqil, Jaida A. Alsuhaibani, Batool A. Alhumaidi, Raghad A. Alnasser, Rahaf D. Alotaibi, Hafida Benhidour
Abstract
Reducing the incidence of gun violence has become a priority nowadays. Using Deep Learning models to automatically detect guns from security camera can contribute in reducing these incidences and save lives. The proposed system in this paper is an Automatic Gun Detection system using Faster R-CNN model. Since it is possible to change the CNN architecture used as a feature extractor in Faster R-CNN, Inception-ResNetV2, ResNet50, VGG16 and MobileNetV2 have been used separately as feature extractors. Intensive experiments have been conducted in order to evaluate the proposed architectures and compare them with YOLOv2. Promising results have been obtained with Faster R-CNN that is using Inception-ResNetV2. However, in terms of training and testing time, YOLOv2 has the shortest time, followed by VGG16, MobileNetV2, ResNet-50, and coming last Inception-ResNetV2.