Digital Watchman Using Deep Learning

Abstract

In response to the increasing instances of fraudulent and offensive activities occurring in real-time, the
need for continuous monitoring of CCTV surveillance footage becomes crucial. Human surveillance alone is not
feasible due to the sheer volume of footage to be analyzed. Furthermore, it is important to quickly identify and
assess frames or sections of the recordings that contain exceptional or suspicious behaviour. This project
addresses these challenges by utilizing various Deep Learning models, specifically Convolutional Neural Networks
(CNN) and Long-term Recurrent Convolutional Networks (LRCN), to detect signs of violence in real-time. The
Deep Learning models are trained on labelled datasets containing examples of violent and nonviolent activities.
By leveraging the power of CNNs, the models can extract relevant features from video frames, enabling the
identification of violent behaviours. LRCNs are employed to capture temporal dependencies and analyse the
sequence of frames, providing a comprehensive understanding of the activities. By implementing these Deep
Learning models, the project aims to provide a rapid and automated method for identifying and flagging
exceptional or suspicious activities. This technology assists security personnel in efficiently monitoring
surveillance footage, enabling them to focus their attention on specific frames or sections of the recordings that
require immediate assessment or intervention. Ultimately, the goal is to enhance security measures, facilitate
timely response, and improve public safety in real-time surveillance environments