Wissensbasierte Mustererkennung zur Erfassung von Bauplänen

Due to technological progress and European standardization (e.g. ISO certification) there is an increasing demand to automate the digital recording of building plans. Currently there are two approaches available: (1) labour-intensive redrawing with digitizing tablets or by screen digitizing, (2) automatic scanning and vectorizing where vectorization generally demands an interactive follow-up treatment due to incomplete or ambigues results. This paper proposes a knowledge-based approach for building plan analysis. The procedure reveals the following processing steps: scanning of the building plan, enhancing the image quality and binarization, extraction of lines, line junctions and character fields, knowledge-based interpretation and grouping of image features to domain specific feature aggregates like door symbols, stair symbols etc., optical character recognition and lexicon-based interpretation of character fields, matching of feature aggregates with dimension sets. First results of a prototype implementation are presented. At last an extension of the approach towards a semantic modeling concept showing a coupling between 3D object modeling and an explict 2D modeling of images and technical drawings is presented