An intensive survey on violence detection from videos using computer vision techniques

Purpose Violence detection is the process of detecting any violent action, including but not limited to punching, kicking, choking, brawls and so on. The usage of video surveillance systems has become prevalent in all parts of the world, and early detection of violent actions would greatly minimize future consequences. This paper aims to detail the various violence detection models which utilize traditional machine learning and deep learning techniques and the performance of the various models have been compared. We have also discussed the challenges involved in the process, the real-time applications, the limitations and the future improvements of these violence detection models. Design/methodology/approach This survey paper describes the techniques used for violence detection from video data, their working and performance, detailing the architecture of the model and its limitations. Findings This survey paper provides a detailed overview of the state-of-the-art research models and technologies in violence detection from videos. As we can see, it has gone from traditional hand-picked features like motion trajectories to features learned using deep learning methods. Key challenges, such as the diversity of violent actions, environmental factors and real-time processing constraints, have been addressed through these solutions. The current trend also includes using the transformer models which are highly successful in natural language processing for this purpose. Originality/value This paper discusses the various approaches proposed for violence detection. To the best of the authors’ knowledge, this is the article that discusses in detail the various violence detection models the challenges and future directions for this model.