A soft computing approach to violence detection in social media for smart cities

In recent years, social media has become an everyday tool for the distribution of videos in which signs of violence appear in different ways. Citizens of smart cities are demanding increasing efforts to authorities in order to maintain public safety, as well as to be efficient in an emergency response. The complexity of monitoring automatically the enormous amount of information generated through social networks results in the need for the development of systems that allow for the automatic detection of violent content in videos. This fact is becoming increasingly important in order to guarantee security for the citizens in any smart city. As a result, this work proposes the development of a system for detecting violence in videos by combining different descriptors that calculate the acceleration produced between two frames of a video. To do this, different techniques, such as the Radon transform or optical flow, are used. The trained system then performs the classification using support vector Machines. The results are promising, with accuracy rates between 85 and 97%, depending on the complexity of the databases used, which demonstrates the validity of our proposal.