Abstract

Abstract:- Current surveillance and control systems still require human supervision and intervention. Thus far, previous workhas mostly focused on weapon-based detection within infrared data for concealed weapons. By contrast, we are particularlyinterested in the rapid detection and identification of weapons from images and surveillance data. We reformulate this detectionproblem into the problem of minimizing false positives and solve it by building the key training data-set guided by the results ofa deep Convolutional Neural Networks(CNN) classier, then assessing the best classification model under two approaches, thesliding window approach and region proposal approach. Hence we have used Faster RCNN, YOLO and Mask RCNN to buildour model. The most promising results are obtained by Mask RCNN based model trained on our database. We also havevalidated our model using loss function to assess the performance of a detection model.