Detection And Recognition Algorithm Research Of Traffic Signs

To ensure the safe operation and efficiency of the Transportation System, in the1980s, the United States, Japan and Western Europe and other countries began to study Intelligent Transportation System (ITS). As an important branch of ITS study field, Traffic Signs Detection and Recognition System provide real-time road sign information to the System or driver by identifying the type and position of Traffic Signs. It has a broad application prospect in the unmanned vehicles (including military unmanned vehicles, the vehicle unattended operation in dangerous situations), intelligent robot, auxiliary driving system.This paper mainly focuses on rapid detection and identification algorithm of the road traffic signs under the moving environment, which is divided into two sections: signs detection and signs identification.Concerning the problem of interest area and traffic sign detection, we presented a traffic sign detection algorithm based on Gaussian color model and SVM. We analyzed the advantages and disadvantages of existing RGB, HSV, YCbCr color space in target detection, and found that the YCbCr color space has good performance in color clustering, and can separate the luminance and chrominance apart. Count the distribution of Cb and Cr color components and express them as two-dimensional gaussian distribution, then build the Gaussian color model, and the coarse segmentation of traffic signs image is obtained. Then the Histogram of Oriented Gradient (HOG) which is not sensitive to illumination changes and scale is used to form target features, we use the feature to train the Support Vector Machine (SVM). Finally, a method based on the combining of HOG feature and SVM is used to further detect the coarse segmented image and determine which shape it belongs to.In order to solve the problem of traffic sign recognition, we discussed several kinds of commonly used classification method in target recognition and carries on the comparison to their identification precision and accuracy, this paper uses the Extreme Learning Machine (ELM) neural network which is simple and training fast for classification. HOG features are chosen as training feature. The dimension reduction of the training sample is done by the Principal Component Analysis (PCA).In this paper, the main work is some exploration and attempt based on the existing sign recognition algorithm. Experimental results show that the algorithm has a better stability and accuracy, also has good effect for bad situation such as partial occlude, light, basically meet the demand of practical application.