A dynamic texture-based method for multi-directional motions segmentation of crowd

A dynamic texture-based method used in crowds´ multi-directional motions segmentation is presented. Firstly, the magnitude and direction of the optical flow field are extracted as the dynamic texture features of the motion crowds; Secondly, initial multi-directional motions segmentation will be performed by means of the dynamic texture features with the aid of the perspective normalization and Renyi´s entropy threshold; Finally, level set algorithm without re-initialization is introduced to optimize the initial segmentation so as to achieve the final multi-directional motions segmentation result. The experimental results show that compared with existing methods, the method presented in this paper is more suitable for the segmentation of the crowd motions with multiple motion directions.