Suspicious Human Activity and Fight Detection using Deep Learning

Abstract

With the increasing number of shootings, knife attacks, terrorist attacks etc. in public places across the world, identifying the wrong behavior of human activities in public places has become an important task. This paper focuses on a deep learning approach to detect suspicious human activity and fight using convolutional neural networks from images and videos. We analyze different CNN architectures and compare their accuracy. We design our systems that can process video footage from cameras in real time and predict whether activity is suspicious or fight found or not. We also propose future developments that can be undertaken to detect and counter distrustful human activity in the public region.