Knowledge-Based Interpretation of Road Maps Based on Symmetrical Skeletons

f' scanner raw.dataS ' 5 The objective of our research reported here is to show the various processing steps on all levels of a computer vision system required to interpret and describe road maps. First, the foundation for the recognition process is laid by computing two sets of distance skeletons. They are termed endoand exo-skeleton and are symmetrical with respect to the boundary between foreground and background of a binary raster image. The skeletons are derived from a novel implementation of Blum's concept of the MAT in the semicontinuum which preserves important features such as Euclidean metric and correct topology. To eliminate noise and quantization effects a new regularization method has been devised based on which the MAT is pruned to its stable inner branches. After the removal of artefacts and further simplifiration of the skeleton the recognition of meaningful complex structures is accomplished in several steps with the aid of a growing amount of domain-specific knowledge. Scene knowledge helps to verify (local) hypotheses in a larger context and to arrive a t a consistent interpretation throughout thr scenr. The selection of the object-oriented programming paradigm proved to be very effective.