This research project is concerned with four basic problems in computer vision and image understanding. (1) Recovery of three-dimensional shape from image intensities. Using photometric stereo and other structural lighting models, a systematic approach to recovery of complete local surface geometry and reconstruction of surface patches from image information will be developed. (2) Recovery of deformation and motion parametersof not necessarily rigid objects. Rigidity has long been a key assumption in computer vision aNd robotics. For motion analysis, this assumption will be relaxed. Many known results will be subsumed. Although there is no uniqueness result in general, closed form solutions will be derived for certain special cases. (3) The correspondence problem using three-dimensional geometric features. Using photometric stereo and other structured lighting models, local and global three-dimensional geometric features will be extracted from objects to establish correspondence of image sequences. (4) Object recognition using range data. Parametric surface patches will be constructed using range data. Efficient algorithms will be developed to calculate geometric features of such surfaces so that they may be used to reconstruct/recognize/learn objects in world space.
