A moving object, the inside of a surrounding field of view of a 2d-daten-feed a calibrated video camera, is recognized and followed by a 3d model by means of locating a centroid of the object and determining an interface with a base plane within the surrounding field of view shown. A suitable network-based grid 3d volume model for the object is by using a rear projection of a corresponding 2 - image as a function of the centroid and the determined interface with the base plane initialized. The nonlinear dynamics of a tracked path of movement of the object is in the form of a collection of different local linear models shown. A structure of the object is projected on the 3d model and 2d-verfolgungen of the object to be expanded 3d - movements, in order to operate the 3d model, in one aspect by learning a weighted combination of the different local linear models, the one image errors due to a renewed projection in relation to a minimized tv-camera.
展开▼