“…Recognition systems generally search for a matching between elements of an object model and instances of those elements in the data, recovering a transformation that maps part of the model onto part of the image. There are a number of different approaches to this model-based recognition problem, including clustering in parameter space (e.g., Stockman [1987], Stockman et al [1982], Thompson and Mundy [1987]), searching a tree of corresponding model and image features (e.g., Grimson [1989aGrimson [ , 1989b Grimson and Lozano-P6rez [1984,1987], Ettinger [1987, 19881, Murray [1987a, 1987b, Murray and Cook [1988], Ayache and Faugeras [1986], Faugeras and Hebert [1986], Ikeuchi [1987]), and directly searching for possible transformations from a model to an image (e.g., Fischler and Bolles [1981], Ullman [1987, 1988]) (see also Chin and Dyer [1986] and Besl and Jain (1985] for more comprehensive reviews). These approaches all share the common property that a decision is ,:ade about the presence or absence of an object on the basis of geometric evidence acquired from the sensory input.…”