Detection And Recognition Algorithm Research Of Road Traffic Signs

With the development of the social economy, the road bears more and more pressure, so the Intelligent Traffic System emerges as the time requires. As an important part of the ITS, Traffic Sign Recognition system (referred to as TSR) is also of concern. The Traffic Sign System generally contains two parts called detection and recognition, mainly using the image acquisition device to get the images for detecting and recognizing. The paper refers lots of relevant literatures and mainly completes the following works through a large number of experiments.First is the color segmentation. Here realizes the segmentation methods in the common color spaces and analyzes their advantages and disadvantages. The R, G and B channels are operated by a nonlinear Gamma transformation to reduce the influence of illumination. This paper puts forward a color enhancement method in RGB and YCbCr space respectively, then uses the threshold to extract traffic sign regions related to red、yellow and blue, also uses the morphological process on each image, including filling empty, eliminating small regions and the regions which ratio of length to width does not conform to traffic signs.The followed step is shape analysis. According to their shapes, the traffic signs are divided into several categories, such as circle, triangle and rectangle. Every region’s edge is extracted for the Fourier descriptor, here uses the first eight coefficients as shape feature. The classifier used is support vector machine (SVM), which is based on statistical learning theory’s VC dimension theory and structural risk minimization principle, mainly for small samples, nonlinear and high dimension pattern recognition problem. This process uses the LIBSVM platform for traffic sign shape classification, also this method can eliminate regions which is not fit for the shape during the color segmentation process.Next is the sign recognition. The paper extracts the regions with the identified color and shape in the original image and sets the parts out of the associated color white so the threshold is easier to get, each two-value image is applied by the principle component analysis (PCA), which can reduce the dimension of the data for easy computation and generally get the best descriptive feature. Then use the BP neural networks for traffic sign recognition, four classes including red-circle、 yellow-triangle、blue-circle and blue-rectangle need to be recognized, each class contains several common signs.Finally the whole system is programmed by the MATLAB software.