Research On Method Of Traffic Signs Recognition Based On Support Vector Machine

As science and technology development and social progress, urban highway congestion, especially the growing phenomenon of frequent traffic accidents, traffic safety has become one of the main issues. So Intelligent Transportation System(ITS)emerged and developed rapidly.ITS is an integrated system that consists of detection, communication, control and computer technology, and etc. It involves the technologies of image processing, digital signal processing, pattern recognition, artificial intelligence, information technology and electronic technology, the communication technology, system engineering technology and so on. Road traffic signs identification system (TSR) is one of the important intelligent transportation system subsystems.This traffic sign recognition system for sign detection and recognition of two stages of the key technologies of a more in-depth research, we design and implement using support vector machine as neural network classifier traffic sign identification system.First, commonly used in traffic sign recognition image preprocessing techniques are summarized, compared and analyzed respectively. According to the characteristics of RGB color model of traffic signs color coarse segmentation, secondly, after using the image of segmentation Sobel operators are edge extraction, and then through the holes filling, area of filtering and shape characteristics, eventually obtaining traffic signs image.Introduces the basic knowledge of traffic signs and the principles of basis, through road traffic signs color - shape characteristic, the design gives a summary of the shape feature based on color - traffic signs image algorithm.Based on the statistical learning theory and support vector machine learning,SVM is proposed based on a traffic sign recognition model. According to the color of traffic signs and geometric properties of correspondence between the shape of traffic signs were extracted features and invariant moments feature as the two neural network input, thus make neural network traffic sign the shape of complete classification and type discriminant functions. This method optimized system flow, reduced the complexity of the system and improve the accuracy.