Colour-based bottom-up saliency for traffic sign detection

On roads, drivers can detect traffic sign extraordinarily fast and accurately; however, the computer vision system does not easily imitate this natural ability although using a lot of image processing techniques. This weakness can be tackled by a bottom-up visual saliency method based on traffic-sign colours which occupy certain ranges of RGB values. These values can be used to modulate the bottom-up visual saliency, so the proposed system can focus on traffic signs detection. The method is tested on the UNMC Automotive Vision Database and compared with results of the purely bottom-up visual saliency method. © 2010 IEEE.