Graefe 1983 : When a computer vision system is used to interpret real-world scenes containing moving objects, certain characteristics of the objects and of the computer vision system itself, that would be irrelevant in the context of sequences of static images, must be represented. Among them are motion-induced blur and distortion of the image. When the scene has to be interpreted in real time, the representation of objects should, in the interest of efficiency, be decomposed in such a way that at each stage of processing a partial representation is available that includes only those aspects of each object that are relevant at that particular stage. Such partial representations may, at the lowest level, relate to the visual appearance of an object, at an intermediate level to coordinates of its parts, and at the highest levels to its dynamic behavior and to error statistics.