Detection of traffic signs based on eigen-color model and saliency model in driver assistance systems

Traffic signs are made in many different shapes and colors that normally stand out from their environment for the purpose of enhancing their visibility to drivers. In most previous research, the methods for detecting traffic signs used shape and color information. However, the shape and color information of traffic signs is sensitive to features of the surroundings, such as the viewpoint and changes in the brightness or background, consequently making such information less reliable in terms of accuracy. To solve the problem, this paper presents a method for detecting traffic signs that is invariant under various environmental changes such as those in the appearance and illumination of traffic scenes. This method uses a saliency model to capture those features of traffic signs that are invariant to shading and shadow. Next, a traffic sign color model is used to extract the color features invariant under illumination changes. By using these models, the presented method can identify all possible shapes and colors for similarly representing the traffic signs. Finally, the traffic signs are detected through a so-called object verification process in which each traffic sign can be recognized by finding a region of overlap between two selected candidate regions. Automatic detection of traffic signs by implementing the presented approach has been examined with traffic scenes of some roads. In particular, the detection rate for traffic signs was 92%, while the processing time per frame was 0.27s.