Knowledge Based System For The Interpretation Of Complex Scenes

Future tasks in remote sensing require automated and robust analysis methods for image data from airborne and satellite sensor platforms. The scenes to be observed exhibit a high degree of complexity. Complexity refers to the large variety of pictorial representations of objects with the same semantic meaning and also to the extensive amount of details of the scenes, i.e. the components which make up a settlement, a rural area, an industrial plant, a street net etc. In order to handle the problem a new approach is presented, which uses structural and holistic methods for object recognition and scene interpretation. The system works on multisensor and multitemporal data. Expectations about the objects and scene content to be recognized can be formulated using different paradigms for knowledge representation. Structural dependencies can be represented explicitly as a semantic net, simple primitives like lines, edges, patches and complex primitives like streets and buildings, etc. can be described and extracted by holistic methods. The system and its components are explained using examples for the recognition of complex patterns like a purification plant, a fairground or the task of the verification of land usage.