A review on crowd analysis of evacuation and abnormality detection based on machine learning systems

Human crowds have become hotspot research, particularly in crowd analysis to ensure human safety. Adaptations of machine learning (ML) approaches, especially deep learning, play a vital role in the applications of evacuation, detection, and prediction pertaining to crowd analysis. Further development in the analysis of crowd is needed to understand human behaviors due to the fast growth of crowd in urban megacities. This article presents a comprehensive review of crowd analysis ML-based systems, where it is categorized with respect to its purposes, viz. crowd evacuation that provides efficient evacuation routes, abnormality detection that could detect the occurrence of any irregular movement or behavior, and crowd prediction that could foresee the occurrence of any possible disasters or predict pedestrian trajectory. Moreover, this article reviews the applied techniques of machine learning with a brief discussion on the used software and simulation platforms. This work also classifies crowd evacuation into data-driven methods and goal-driven learning methods that have attracted significant attention due to their potential to adopt virtual agents with learning capabilities. This review finds that convolutional neural networks and recurrent neural networks have shown superiority in abnormality detection and prediction, whereas deep reinforcement learning has shown potential performance in the development of human level capacities of reasoning. These three methods contribute to the modeling and understanding of pedestrian behavior and will enhance further development in crowd analysis to ensure human safety. © 2022, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.