Research And Implementation Of Multi-features Traffic Sign Recognition Based On Support Vector Machine

With the improvement of people’s living standard, the car gradually become popular, more and more become a popular means of transportation. And the subsequent, much attention has been paid to traffic safety problem. The incidence of traffic accidents has increased every year, despite the traffic regulations constraints and road traffic signs remind and warn us, it seems not to be able to be vigilant so that the driver pays attention to traffic safety. With the auxiliary driving, unmanned and the concept of road Traffic Sign automatic detection and so on coming up with, it makes the Traffic Sign Recognition (TSR) researches have a very important practical significance and practical application value.Traffic sign recognition mainly involves two aspects:traffic sign detection and traffic signs content recognition. In view of the obvious color feature and the shape feature of traffic sign, at the detection phase the color as the main basis for image enhancement is a good choice, and the calculated dual-threshold as the binary threshold for binary image has a good result, then we choose the experience threshold combined with the adaptive threshold as the area threshold selection method to segment the interested area for further selection, if there is an area of relatively large proportional requirements of its area division, then we will estimate whether the internal of the area contains traffic signs. Classification phase, the training and study is mainly based on support vector machine. The first step, we will distinguish the shape of the traffic signs, in the shape of edge profile by edge direction histogram of feature extraction, and then use SVM to classify; The second step, according to Hu moments and Zernike moments, unchanged from the internal region of traffic signs information description, again through the training of support vector machine (SVM) to determine the accurate information of traffic signs. In the process of the shape of the support vector machine (SVM) classification, specific to the edge contour weighted characteristic value allocation; find out the defining features of each shape, weight distribution, so as to improve the recognition rate; in the process of the area information of SVM classification, according to the constructed by the lower order moments of Hu and Zernike moments invariants constructed by the higher order moments to describe the details of the image, so as to achieve the ideal recognition of the traffic signs’contents.The experimental results show that, based on the defining feature selection and distribution of value, to a great extent, improved the recognition and classification accuracy of support vector machine (SVM), which provides the inner region of indirect information discriminated accuracy. Hu invariant moments and Zernike moments combined together as traffic signs identifying information also provides a good support to improve the recognition rate of SVM secondary content. The experimental results show that the algorithm has a good accuracy and robustness.