Traffic Sign Recognition based on Color Standardization

It is known that the key factor affecting recognition rate of traffic sign is the color distortion caused by the color complexity. By analyzing the color characteristics and classification features of traffic signs, this paper proposes a novel model of the traffic sign classification method using color standardization based on the rule of "simplifying the problem, using the color information and the intelligence approach". The first step is to reduce the complexity of the color information to five types of standard colors, and then employ two-layer intelligence classifier for the classification. Because support vector machine (SVM) can provide the good generalization and over-fitting avoidance performances, the SVM network is applied for color standardization and the traffic sign classification. Here, we first carry out preprocessing on the traffic signs, and this step is called as color standardization, in which the color specification library is constructed with the colors: red * yellow * blue * white and black. 16,777,216 kinds of color made of 24-bit bitmap are mapped into a single space of five elements, which significantly simplifies the complexity of the traffic signs' color information and is more suitable for the traffic sign classification. Secondly, two-hierarchy intelligence classifier is employed for the classification of the color standardized traffic signs. 23 standardized ideal signs are selected as training set, 531 standardized real signs as testing set. To both sets, the SVM network is applied. By doing so, 100% correct recognition rate in rough classification and 70% average correct recognition rate in fine classification are achieved. The proposed approach has also shown better robustness for adapting to the variation of the color distortion, the structure parameter and the training network parameter.