Detection of Fake and Real Violence Using Hierarchical CNN Model

This investigation delves into the intersection of deep learning and image processing for early detection and classification of violence, with a primary focus on differentiating between movie fights (staged or fake) and true violence. Leveraging the "Violence and Non-violence Images Dataset," along with the collected movie fight images dataset, the proposed methodology involves Training Model3 (Hierarchal combination of Model1 and Model2). The hierarchy enhances performance and significantly improves specificity scores, even in a dataset biased toward nonviolence cases. The proposed model achieves an impressive accuracy of 98.33%, showcasing its potential for crime detection.