Violence Detection in Real Life Videos using Convolutional Neural Network
Nandini Bagga, Gajan Singh, Balamurugan Balusamy, Ajay Shanker Singh · 2022 2nd International Conference on Advance Computing and Innovative Technologies in Engineering (ICACITE) · 2022
Detecting violence in video recordings through artificial intelligence is critical. Just because violence detection seems complex now, it doesn't mean that the problem should be ignored. With this paper, we aimed to show how we can do violence detection and how it can be implemented easily with the simplest methods at hand thanks to the progressing advances in deep learning and AI. Furthermore, it could be an excellent tool for preventing children from accessing inappropriate content and assisting guardians in making better decisions about what their children should watch. This is a challenging topic since the definition of violence is broad and quite abstract. As a result, recognizing such intricacies from recordings without human intervention is not just a technical issue, but also a theoretical one. In light of this, using a convolutional neural network, we will investigate how to depict the concept of violence. For image categorization, there are numerous pre-trained convolutional neural networks available. These models have figured out how to bring out dynamic and useful highlights from regular pictures and use them as a beginning stage to get familiar with new tasks. These networks can classify images into various object categories after being trained on over a million photos. We used pre-trained networks with transfer learning since it's regularly much quicker and more straightforward than preparing or training a network from scratch. Using several deep learning techniques, we will investigate which model will provide the best accuracy for recognizing violence in videos for this project.