Real time road sign detection based on rotational center voting and shape analysis

We present a real time road sign detection framework based on color component extraction, rotational center voting and shape analysis. The color component extraction comes from so called color double-opponent in human primary visual cortex in which one color is excited and another is inhibited. For the rotational center voting, we use the pairwise gradient vectors vote for their rotational symmetry centers by which centers and scales of regular polygons can be detected. Meanwhile the points which voting to the centers will be recorded and the categories of the sign shapes can be decided by analyzing the points. The method is tested on Chinese road sign dataset which is collected ourselves and also on the UHA dataset used by many other researchers. The experiment shows that the proposed method is invariant to translation, scale, rotation and partial occlusions.