Knowledge Based Modelling of Natural Scenes

A knowledge based approach for the automatic generation of 3D models from stereo images is presented. Two different application fields are shown: The modelling of houses and the modelling of landscapes from aerial images. The models are used for computer graphic applications. A human observer is in possession of a priori knowledge about the scene. To meet his expectations and thus to improve realism the presented approach employs explicit knowledge. The knowledge base is exploited to derive constraints for surface reconstruction and to complete partially occluded models.