Violence detection in videos using 3D convolutional neural networks and transfer learning

Automatic violence detection using computerized systems instead of manpower has been a subject of significant contemporary interest among researchers recently. In addition, Deep Learning models such as Convolutional Neural Networks have been successfully applied to many different tasks in a wide range of domains, including video recognition. To that end in this thesis, a computerized model for violence recognition will be designed which does not require manual human inspection. Two models will be designed, including a simple 3D CNN and a MoViNet 3D CNN which uses transfer learning. A combined dataset consisting of 5200 videos will be used to train and run the models. The aim of this thesis is to give a comprehensive explanation to the design and mathematics of CNNs, implement two 3D CNN models and explain and analyze them in many aspects.