Traffic Sign Recognition Based on Pictogram Contours

This paper addresses the problem of automatic recognition of traffic signs from images captured while driving, to provide a driver aid. Traffic signs can be detected based on their color and classified, also according to their shape, into danger, information, obligation or prohibition classes - special cases are the stop, yield and wrong way signs. Then, traffic signs can be recognized based on their pictograms, as each pictogram is unique within a given sign class. The proposed recognition algorithm analyses pictogram outer contours. When pictograms are composed of several disconnected regions, a snake is used to create a single pictogram outer contour. Matching of signs against the database is done using the curvature space scale representation.Several examples taken from Portuguese roads are used to demonstrate the effectiveness of the system.