Moving human detection and extraction via improved optical flow and adjacent region merger

Human action recognition plays a significant role in intelligent video surveillance system, of which moving human detection and extraction are the most basic and critical parts. A novel improved optical flow and moving human extraction method are presented under the fixed camera in this paper. We improve the optical flow with inter-frame difference method and optimized gradient algorithm, which can not only increase the accuracy evidently, but also reduce computation time effectively. In order to more precisely extract the moving human, this paper also proposes a method to obtain moving human bounding box on the basis of Adjacent Region Merger (ARM). By comparison with other optical flow methods, the superiority of our algorithms is further proved.