Videos displaying violence or sensitive content are harmful and can negatively impact a person’s mental health. Videos like this on social media should be marked with a trigger warning or censored for individuals who don’t like to view violent content.
In this project, you can build a deep learning model that detects violence in videos and automatically generates a warning, informing users to watch it at their own risk.
To train this model, you can use datasets that contain both violent and non-violent content (these will be linked below). You can extract image frames from these videos and train a CNN on them. There are various pre-trained models that you can use to accomplish this task, including VGG16, VGG19, and Resnet50.
People have managed to achieve high accuracy scores (over 90%) for this task with the help of transfer learning. Since transfer learning uses models that have already been trained on millions of general images, these models usually perform better than models you train from scratch.
Datasets: Violent Flows Dataset / Hockey Fight Videos Dataset
