This research is concerned with answering two questions: how can one robustly infer the boundaries of grouped data sets and how can one characterize the relationships between the subsequently inferred models? A robust model inference is achieved by an aggregation of data (range measurements) into shape models (or parts) that considers both local and regional geometric information. That is, robustness is achieved by the design of a cooperative process that integrates local shape cues (negative minima of curvature) and the consistency of the region with its associated representation (superellipsoids). Part boundaries are determined by integrating model parameter estimation with curvature extrema detection.
To completely characterize the significance of the scene via a part representation one must explicitly represent both the intrusions and extrusions (the convexities and concavities - the figure and ground) that determine form. That is, our shape primitive will model not only the solid shape itself but also the environment surrounding the form. The research has investigated how one can describe a scene of multiple objects and their constituent parts by examining the relationships between the properties (for example, orientation, size, curvature, etc.) of the inferred models.
T. Jelonek, F.P. Ferrie