Symmetriesignaturen für bildbasierte Anwendungen in der Robotik

The robots that are to find their way in our future households and everyday lives are necessarily have to be mobile and self-dependent. For such autonomous systems it becomes more and more important to efficiently process the incoming data and to thereby eradiate what we might call intelligent behaviour. While intelligence in terms of plans and goals are abstract metaphors of each robots decision process, the perception of the local environment has to be a central issue. Dealing with this demand in the context of intelligent systems shows plainly what sophisticated human visual perception is like. The creators and developers of artificial systems therefore build up a construction kit with construction blocks that try to represent the reproduction of cognitive perception mechanisms by machine algorithms. In this work, a construction block for symmetry is added to this set. The main chapters discuss three layers that describe this block from its origin of biological motivation up to its application for intelligent systems: The first layer addresses the basic motivation of symmetry as a feature. Therefore, symmetry references are given in diverse domains such as psychophysics, biology, mathematics and computer vision. New methods are developed which provide description and application of symmetry as an image feature. The second layer proceeds to application in relation to state-of-the-art modern regional image features. The three steps of detection, description and robust matching of regional symmetry features form the necessary links between the basic motivation and the practical application of symmetry. Symmetry features are evaluated and compared to state-of-the-art features considering their robustness w. r. t. common image transformations. Practical examples from the area of mobile robot navigation are proposed in the third layer. These show the capability of the developed symmetry features in applications. They also provide the links to higher-level construction blocks from the set of visual object analysis and robot navigation. The main contribution of this work is a fundamental analysis of symmetry as an image feature, the development of a robust visual symmetry feature detector and the implementation of symmetry in robot applications. This thesis is written in German, except for a concluding section at its end and this introductory summary. After the concluding section, the interested reader will find references to several international publications that comprise the main aspects of this thesis in English.