Study On Traffic Sign Automatic Identification Technology Of No One Car

With the development of economy and the improvement of people’s living standard, in today’s society more and more cars got popularity, moreover the new situation of the current energy environment and climate change. In order to solve the above-mentioned problems, in the early1980s Japan and many European countries began to research the intelligent transportation system(ITS), to promote the development of traffic and its coordination with environment. Traffic sign recognition is a key technique in the construction of intelligent transportation system, although in the initial stage of intelligent traffic system construction, some relevant research have begun, However we still don’t have a very complete and mature traffic sign recognition system. The research of traffic sign recognition technology started quite late in our country, for efficient and practical traffic sign automatic identification technology still need further research.Firstly this paper introduces the domestic and overseas development of UGV’s technology and traffic sign recognition technology, and analyzed the gap in both at home and abroad. Secondly, studies the image preprocessing methods, color space model method and feature extraction algorithm. The traffic sign collected in the natural environment is easily affected by uneven illumination and caused the image distortion question.For this problem has carried on the image enhancement。 It can be seen that image distortion problem got very good settlement through the experimental treatment effect. On the basis of RGB color space model component fixed difference value introducing the color on traffic sign segmentation,and though the use of open operation,closing operation, corrosion, expansion algorithm for segmentation of traffic signs to noise processing. Then, according to the situation of the divised traffic signs image information is uncompleted, Using SIFT algorithm and Hu invariant moment respectively on traffic signs for feature extraction. Due to SIFT features higher dimension, the recognition of real-time can be affected. In this paper,on the extraction of SIFT feature dimension reduction by using principal component analysis. Through the experiment proved that PCA-SIFT descriptor generation time is the traditional SIFT descriptor generation time decrease greatly, improve the real-time of feature extraction. Finally, this paper studies the basic principle of Bag of Words model and support vector machine. Based on dimensionality reduction after the SIFT features in combination Bag of Words model, and then puts forward the suitable for traffic sign recognition practical model. In order to realize the good recognition effect and generalization performance of road traffic classifier, We performed classification through structure Multi class classifier based on support vector machine. A lot of experimental results show that the proposed Bag of Words model algorithm has characteristics of high accuracy and good real-time.