Representation of deformable object structure and motion for autonomous manipulation using relative elasticity

Even though most natural and many man made objects move in a nonrigid manner, the majority of the research in computer vision has focused on recovering motion and structure of rigid objects due to the underconstrained and complex nature of deformable transformations. A novel image based model for the representation of nonrigid structure and motion based on relative elasticity is discussed which provides both sectional and global motion information and requires no knowledge of physical parameters, allowing it to overcome many of the limitations of existing deformable models. The usefulness of this new representation is demonstrated for autonomous manipulation of nonrigid objects