A Fast Crowd Segmentation Method

As a hot research interest of computer vision field, crowd behavior analysis has broad application prospects and great economic value in many aspects, such as intelligent video surveillance, virtual reality, public safety and so on. In high density crowded scenes, pedestrians can be seen as a collection of several groups which consist of people sharing similar motion features. Groups are necessary parts that make up a crowd. To obtain basic crowd interactions for next crowd behavior analysis, a fast crowd segmentation method based on spatiotemporal relationships of crowd is proposed. This method not only considers crowd motion information but also spatial information of pedestrians. The method can achieve the same segmentation results by using less time and can be applied to various crowd scenes. Extensive experiment results on videos of real-world crowd scenes demonstrate that the proposed method is effective as well as efficient for crowd segmentation.