Detection And Recognition Algorithm Research Of Traffic Signs In Natural Environments

In recent years, more and more researchers pay attention to the IntelligentTransportation Systems (ITS).Traffic Signs Recognition is an importment part ofIntelligent Transportation Systems. It has very important theoretical significance andpractical value. But as the complexity of the natural environment, detection andrecognition of traffic signs is not mature. It needs us to research deeply withpractical efficiency of detection and recognition of traffic signs algorithms in thisfiled.The purpose of this study was designed algorithm to detect the images thatcaptured in the natural environment, detected the area that included traffic signs, andidentified the detected traffic signs, The paper conducted the following research:（1）we analysised the standard traffic signs, choiced the color, the inner color,the shape and the outer contour as the feature of traffic signs.（2）Image preprocessing, weaken the interference of background,lighting andother factors.Comparaed and analysisd the enhancement algorithms of the colorchange and color constant. Comparaed the enhancement algorithms of doubleGamma base on algorithms of the color change and the enhancement algorithms ofthe average of local color base on algorithms of the color constant.（3）We tested images and found out the area with traffic signs. First of all, wemeasured different color of traffic signs in HSV color space. As the images’background are complex, color is part of the detected impurities,we need to removeimpurity of every possible area. We used pixel expansion to mark every possiblearea different values achieve ROI segmentation. Confront the different sharp, color,outer contour.We use center projection, radon transform and internal color judgmentfor each part for further testing. Through experiment We found the comprehensiveuse of various detection algorithm can correctly detect contains the region of trafficsigns.（4）We use the support vector machine (SVM) to testd and classified thedetected traffic signs. After binarization of the traffic signs,we use Hu invariantmoment and Zernike invariant moment feature calculation for different categories of feature vector. We selectd red traffic signs as object to make a experiment. And weused physical objects in the collection of traffic signs as the training and testingsamples.At last, we used the LIBSVM traind samples and testd experiments todetectd the traffic signs. Through the experiment found a better recognition rate.