Abstract
Due to the rapidly increasing number of requests from the surveillance and security industries, detecting a panic behavior in human crowd have become an active research in the field of computer vision in recent years. The automatic and real-time analysis of video sequences in order to detect a panic behavior is one of the most challenging tasks for computer vision experts. This stems from the fact that a panic is mostly defined as a sudden change in the pedestrian movements and that motion estimation is computationally expensive. The aim of this paper is to give an overview of the reported real-time techniques in the literature by focusing on how these techniques overcame the heavy computations of the motion estimation. Moreover, the present work summarizes the accuracy and the execution time of those techniques to highlight the direction to future studies.