Incremental recovery of volumetric models

This research investigates the use of a deformable model representation in the context of an active exploration process. The exploration process is used to acquire data, fuse measurements from different viewpoints, and construct shape models. Until now, approximate shape models, e.g. superellipsoids, have been used to represent objects in an active exploration strategy. Although these models provide an adequate characterization of 3D shape for many tasks, a more detailed representation is needed when it comes to tasks that require discrimination of similar objects.In this thesis, we explore the use of a class of deformable models, developed by Pentland and Sclaroff, to further refine the initial coarse superellipsoid description to provide a richer description of the scene. The autonomous exploration strategy used is the one developed by Whaite and Ferrie. Along the way a number of specific issues will be addressed, namely how the models are initialized, the stability of the model parameters throughout the exploration process, and the attachment of data points to the model surface. This work demonstrates the utility of the deformable models for shape interpolation and as general representations for 3D shape