Abstract
This Paper describes an action recognition method from motion pictures picked up by a single TV camera. In This method 2D-Feature points such as a head top, shoulder edges, hands and so on are extracted and matched with the reference data of a 3D model.Under the hypothesis that an object person stands on a floor, the position of the above feature points of an object person can be estimated by a triangulation. Through these processes ,the human actions can be analyzed.
