Study On The Algorithm Of Road Traffic Sign Recognition

With the progress of technology and the development of social economy, automobile has become an indispensable travel tool in the society. Road Traffic also has become increasingly complex, road safety has become the focus of the society pay attention to generally. Traffic sign is distinctly the important part of Road Traffic, traffic sign makes auto have right way, cues driver to operate and makes traffic safer. It could help drivers get the message about around environment, such as: no-stop,no-pass,no-ring,no-left turn,no- right turn,and danger warning .It reduces a lot of unnecessary traffic accidents, owing to traffic signs. However, there are still many traffic signs ignored by some drivers, or because of some special circumstances in traffic accidents. In order to reduce traffic accidents by these reasons, many scholars and experts have begun traffic signs recognition. Traffic sign recognition has become an important part of the semi-automatic or automatic vehicle.Traffic signs recognition system which being an important part of intelligent traffic system recognizes traffic sign type in front of the vehicle by vision system and defends warning for the driver in real-time. So studying on road traffic signs recognition system is necessary and practically significant.Traffic signs have their own characteristics and recognition algorithm in accordance with their characteristics can be studied. The color and shape of traffic signs are mostly notable features, this paper studies the traffic signs recognition algorithms through the two characteristics and three kinds of traffic signs - Prohibition of left turn, No-long or temporarily parking and No-access.These three signs have the same characteristics--red color contour, white background and prohibiting left turn sign has Black kernel; so I research traffic sign recognition algorithm according to these characteristics.In the paper, the traffic sign recognition algorithm should have reliability,validity and real-time requirements. According to the result of the experiment research on algorithms, we can see that the algorithm is very valid, and amount of computation is very small. The main contents of the paper include: 1. Histogram equalization color image enhancement algorithms based on I(intensity) channel of HIS color model space ; 2. The threshold segmentation algorithm based on RGB color model space;3. Based on the characteristics of the circular shape of the objects to extract the target region;4. The edge detection algorithm based on Log operator and the thinning method of mathematical morphology to extract single-pixel edge; 5.The traffic sign recognition of features matching based on improved Hu unchanged moment and affine unchanged moment.1. It causes photo's color distortion, due to weather conditions, traffic information, or the reasons for the vision sensor. The paper used histogram equalization enhancement algorithms based on I (intensity) channel of HIS color model space, and histogram equalization algorithm is the principle of the average gray level of the picture, the picture tends to average brightness, the color of the images is closer to ideal.This paper uses this algorithms, so the target areas of the images are color enhanced.2. According to the color characteristics of the objects, the paper uses the threshold segmentation algorithm in RGB color space and in H (hue) channel of HIS color space respectively, and the object area and the background area will be separated into the binary images. In RGB space, it makes objects separated from the background distinctly, and saves time; while there are many noises which are similar with the color of objects in the images. Segment the image in H channel of HIS space, it can segment red very distinctly and noises are very few, but black has not hue , it can't segment black objects area from the background.However,because there is a course of RGB to HIS space conversion process. By comparing the analysis, this paper uses the RGB color space segmentation. Because of noises, the paper uses isolated point algorithm to eliminate single points and threshold area algorithm to eliminate some big areas which are not the object regions.3. According to the circle characteristics of the traffic signs of the paper studying, the paper utilizes circle classical characteristics—the circle algorithm to extract targets which may be. The calculation of the circle is obtained by the size and perimeter of each region. According as the threshold of the circle we advanced set off to eliminate noises which can't be the objects.The circle is simple in principle, has a small count of calculation, good real-time and has good extracting effect.4. Extracted circular areaes, the image needs to be done further treatment to prepare for the identification work.The paper utilizes the expansion of mathematical morphology which makes objects smooth to eliminate burr. Then the paper utilizes the edge detection operator to extract the edges of the objects, we choose Log operator to extract edges of the objects by analyzing the effects of several classical operators. It is necessary to make edges single pixel for the next matching algorithm. We uses Thinning algorithm of mathematical morphology to make the objecs become single pixel edges.5. Recognition is the most important work in the paper, in order to meeting rotation, translation, scale invariance of the images, the paper uses improved Hu unchanged moment and affine unchanged moment to match, and through comparison ,the paper selects the most representative six unchanged moment as characteristics vector. According to characteristic feature vector template, and calculate continental distance between characteristics templates and the characteristics of the images, then make judgments contrast threshold and identify the categories of the targets.