Model-based object recognition is a popular and difficult subfield of computer vision research. The problem of model-based recognition of free-form objects is addressed in this project. This problem has many practical ramifications and robust solutions will be applicable in areas including modern manufacturing and autonomous navigation. Prior work in model-based recognition will be used to develop representations for 3D objects which are suitable for use in recognition, prototype recognition strategies with an explicit goal of efficiency, and disseminate their research results in the standard forms, including electronic archives available over the Internet. The project has the following major components: (1) Development of an object representation strategy for free-form objects suitable for both image synthesis and object recognition. Point-based representations are considered along with surface triangulations. (2) Development of a recognition and localization technique based on the representation strategy chosen. An industrial collaboration will allow the PI to obtain variable-resolution triangulation-based models of free-form objects from real range data obtained from real sensors. A small tabletop three-axis mill will be used to fabricate prototypes of free-form objects from CAD designs.
