A Review of Violence and Riot Detection Techniques

Abstract

The world is changing every now and then at the speed of human thoughts arises in the mind. These thoughts can be shared on Internet and social media platforms and it may connect other people to the same thought. Whether it is the Internet world or Physical world, Security is vital and needs to be addressed carefully. In the Physical world CCTV Surveillance is an important way of security. It captures real time footage of the area that is under threat, whenever there is any unwanted activity under that area then that footage needs to be analyzed for detecting unwanted activities. Nowadays CCTV cameras are a prime source of securing any area that is under threat, so there is an increase in the number of Surveillance cameras so the size of recording also increases manifold. Now the challenge is how this Surveillance footage will be analyzed for timely action. Detecting unwanted activities in video and images with the help of some automated system is the need of today's challenges of security. Machine Learning or Deep Learning may help us in detecting these unwanted activities in video and images in real time for sending timely notification and taking action against these activities. Action recognition is a broader aspect of violence detection that needs great effort and data. However my work is limited only to recognize violence activities like fighting, rioting, stone pelting and violence done by a group of people or a riot-like situation. My whole motivation towards this work is because of 23 Feb 2020 Delhi Riots. These riots compelled me to find some technical solution to prevent and detect these riots. The intention of this research work is to study different methods and models to predict violence like behavior using Machine Learning or Deep Learning in video and images. This Paper discusses deep learning methods like Support vector machine (SVM), Convolution neural network (CNN) and recurrent neural network (RNN) to detect violence activities. Additionally, the video features and datasets that are used in the algorithms and are crucial to the recognition process are also covered. For better comprehension, the research methodologies' steps have been outlined here. The overall research results that may be useful for identifying prospective future work in this research field have been discussed