Video-Based Recognition of Walking States

An approach to analyzing human walking states based on multi-viewing-angle was proposed in this paper. Firstly, the moving person was detected using self-adaptive background updating algorithm. Then, the moving period of the centroid in the upper part of the human body was extracted by 4-direction chain code and the next motion of person was predicted. Finally, fuzzy set of walking states was obtained by the training set of period, and three kinds of walking states including promenade, walking and running were analyzed by membership function. The experimental results show that the movement velocity and the walking states are exactly analyzed by the proposed method, and furthermore, the method is reasonably robust in background noise and varying view angles.