Deeply Supervised Practical Implementation of Violence Detection from Videos for Maximizing Performance

 The purpose of violence detection is to determine whether or not a  violent action has occurred. This topic grew in popularity due to the need to develop appropriate and automatic violence detection systems that investigated visual data obtained from security cameras positioned in various regions. To classify the videos into four different crime categories or non-violent, we downloaded 200 videos from multiple sources such as news channel recordings, YouTube, and online posted videos on all other social networking sites. We created four different classes, such as arrest, assault, arson, and abuse, with each class containing 50 videos. This work used pre-trained deep neural networks to provide a low-complexity strategy for detecting violence.