This paper proposes an improved color barycenter model (CBM) for road sign detection. The previous version of CBM can find out the colors of road-sign (RS), but its accuracy is not high enough for magenta and blue region segmentation. The improved CBM extends the barycenter distribution to cylinder coordinate and takes the number of colors in every point into account. Then the K-meansclusteringisusedtoanalyze the distribution under cylinder coordinate. Using Geodesic distance instead of Euclidean distance for K-means clustering and some conditions provided by the initial color region of CBM is used to constrain K-means operation. The experimental results show that the improved method is able to detect RS with high robustness.