Graphic Recognition And Understanding Of Unmanne Vehicle Traffic Signs

As the rapid development of economy, the vehicle has been widely used, which is a essential traffic tool now, and also followed by more traffic accidents. Traffic sign recognition, so the ITS (intelligent Traffic System) has gained more and more attention. As an important part of the ITS, the RTS (Recognition of Traffic Signs) has become a research hotspot. However, it’s a challenge because of some changeable factors in the actual environment, such as varying illumination. So this thesis prepares to do some algorithm research on detection, correction, texture feature extraction and traffic sign recognition. The main content and results are as follows:(1) This thesis combined with RGB dynamic difference and shape information, to realize traffic sign location. Firstly,in the first location based on color, compares the color segmentation approach in RGB space, HSV space,considering the requirement of real-time, we use a RGB difference method, which realizes the color segmentation by extracting the RGB dynamic difference, we extract the traffic sigh area according to the difference of each component in RGB space, and to mean the pixel in the target area. So, a new difference is obtained based on the average value. Finally, this paper finishes the traffic sign segmentation. Secondly in terms of precision location based on shape, analytics the shape of the three kinds of traffic signs and design algorithm, use deformation approximation algorithm to location the triangle signs. experimental results show achieves the good localization effect in complex environment.(2) This thesis introduces a traffic sign correction method based on sparse and low rank. Traffic signs internal texture have deterministic and regular structures, based on this feature, In view of the internal texture of traffic sign, sparse low-rank approximation and affine transformation are combined. we set the affine transformation matrix of x, y direction and calculate the optimal rank of low rank, which is judged by the soft threshold computation. when the rank is optimal, the corresponding image texture represents the final correction of traffic sign. Experiments showed that this method has strong robustness, and improve the next recognition rate.(3) This thesis proposes to extract the traffic sign kernel based on saliency, which the histogram contrast is used, because the quality of texture directly affects the final recognition rate. Firstly, computing each pixel’s distance metric of R component in RGB color space, we obtain the saliency value of every pixel. Then, smooth processing is followed, at last, Otsu is used to get a adaptive threshold and generate the binary image. Experiments demonstrate the proposed method improves the final recognition rate.