Road sign recognition is a driving assistant function in foreseeable future. This function produces not only the detection of road signs but also the interoperation the content of the signs. In order to obtain the most reliable result in a short processing time, one of the strategies is to find the most suitable image of the sign for the optical character recognition (OCR). This paper proposes an automated procedure to detect road signs, and to track them until the image of the best quality is found. We used multi-scale colour clustering procedure to find road signs, apply a linear motion based tracking procedure to increase the robustness of the detection, and propose a statistic measurement combining all the colour “behaviour” within the road sign region to indicate the quality of the road sign image. The proposed procedure is tested on the image sequences obtained from British and French motorways.