Physical violence detection using a deep-learning model

Abstract

Physical violence-related crimes are rampant in Uganda and the rest of the world. These include; muggings, beatings, robberies, assaults and so many others. As a result, citizens feel extremely insecure and fear night movement. To make matters worse, some of these violent crimes occur in broad daylight. This threatens the citizens’ freedom of movement in pursuit of a better way of life. It also instills fear among the people which slows down economic development. The police force and the Army have done their best to install millions of CCTV cameras to improve surveillance in the country. Patrols are done regularly by the army and police forces to keep law and order. And yet the crime rate is still on the rise. In this research, we designed a Convolutional Neural Network model, which is a deep-learning AI model that recognizes and detects physical violence in an uploaded video. The model was deployed to the cloud, and can be accessed through an API, therefore can be accessed from any platform(mobile, web, desktop). We then built a mobile application to test and validate the physical violence detection model. The mobile app is able to detect physical violence in an uploaded video with a high degree of accuracy. This can, later on, be applied to CCTV cameras to help detect violence quickly and efficiently by different entities such as the police and private CCTV camera owners.