The purpose of this research project is to develop new theories and techniques which can lead to computer vision systems having the capability for "generic object recognition" (that is, identification of objects at the category level). This capability is a fundamental part of the long-term research goal of visual perception for an autonomous real-world systems. The approach to generic recognition that is being pursed in this proposal is termed form and function. The basis of the approach is that concepts of object categories are naturally formed in terms of object function, and that reasoning about object "form" (3D shape) can be used to determine the possible "function" of the object. A recognition system has been implemented which successfully demonstrates this approach for polyhedral shape descriptions and the object category chair. This initial success demonstrates that the basic approach is valid and worth continued exploration. This project will result in extensions of the theory in the areas of (1) incorporating multiple distinct super- ordinate object categories, (2) function-based modeling of object categories which include articulated objects, and (3) analysis of shape descriptions acquired from real data.
