Structure from limited motion of complex rigid objects

A feature-based method is presented for recovering the structure of rigid objects rotating about a fixed axis. The emphasis is on using several views that span only a small angle of rotation to recover the structure of a complex object. First, the correspondence problem for densely-packed features of a complex object undergoing slow rotations is solved by applying the smoothness of motion constraint to higher-order motion approximations. Second, the inaccurate feature motion solution that results from having only a limited extent of rotation is augmented by utilizing any symmetries discovered in the object to constrain one feature motion solution by the gross approximate motion of another feature. These constraints are in addition to those derived from the assumptions of a fixed axis of rotation and a rigid object. The data source presented is a sequence of radar range-Doppler images; similarities to optical image understanding are noted.<>