Traac Sign Recognition Based on Color Image Evaluation 1

This paper originates from a cooperation of Daimler-Benz AG, the University of Paderborn, and our group, within the European PROMETHEUS project. Common aim is the development of a traf-c sign recognition (TSR) system. The Paderborn group provides gray-level image evaluation, Daimler-Benz AG pictogram identiication, and our group color image evaluation. Our approach to TSR is based on a fast and stable color segmentation, the CSC (Color Structure Code). For the segmentation we use a newly developed hierarchical region growing method. This technique combines the advantages of local (simplicity and quickness) and global region growing methods (robustness and accuracy) and avoids chaining mis-matches. The result of this segmentation is a hierarchical data structure. In the evaluation phase geometrical primitives (as circles, ellipses, triangles, rectangles , etc.) can be quickly found in this data forest. Answering a few and simple questions (about inclusion of certain colored geometrical primitives in others) provides a fast but nevertheless surprisingly good detection and classiication of traac signs.