CONTRIBUTIONS TO A 3-D ROBOT VISION SYSTEM - GROUPING FROM SPARSE ANDINCOMPLETE DATA

This thesis considers intermediate level vision, which is the most und erdeveloped domain in computer vision. The task of intermediate level vision is to find sets of composed entities, which are structured comb inations of the simple features in image data. This process is often c alled grouping: generate order from chaos by detecting structures of s ingle physical objects or their parts. Structural grouping of informat ion plays a fundamental role for both computer vision and human visual systems. New methods for grouping sparse and incomplete data for find ing descriptions of shape are explored in this thesis. The major issue s for developing new methods has been the need to achieve computationa l efficiency and robustness. The main motivation has been to provide m ethods for a model based, 3-D object recognition system. This visual r ecognition system was used for controlling a robot arm, which was capa ble of manipulating complex, three-dimensional, real-world objects clo se to real time. One part of this thesis focuses or adaptive setting o f the poll size in Probabilistic Hough Transforms from sparse data. It is experimentally demonstrated, that the suggested methods for adapti ve stopping rules call for polls with average size that is lower than the fixed poll size that would lead to the same error rate. Another pa rt of this thesis introduces grouping processes for finding axial desc riptions of symmetrical shapes from incomplete edge data. It is illust rated, that the detected shape descriptions are useful for feature ext raction, object recognition, shape description, and stereo corresponde nce.