Real Time CCTV Violence Detection System Using Deep Learning

Abstract

Violence has been a menace in every city across the world but with the increase in surveillance cameras, it has become easy to keep check on it. Though it is not possible to physically monitor all the locations at once. In this paper, we intend to create a system to raise an alarm in situations of violence for quick intervention of the local police to avert any mishaps and serious escalations. In this paper, we have proposed using a CNN based MobileNetV2 for feature extraction because MobileNetV2 is known to provide a higher accuracy and also has a lower response time after which a deep learning algorithm of LSTM is used. There are various datasets available for violence detection but we aim to create our own dataset for the model The dataset used is created using a collection of videos on violence from various social sites like Youtube, Shareavideo, Mediadrop. Another dataset is used for determining the accuracy of the model further.