In the face of increasing traffic jams and serious accidents, intelligent transportation system (ITS) and the technology of unmanned vehicle are raised in response to the proper time and conditions. As an important part of ITS, traffic signs detection and recognition system (TSDRS) has become the hot topic for the research. In generally, TSDRS contains two important processes:one is detecting and locating the position of traffic signs, the other is recognizing the meaning of traffic signs. This issue mainly includes following aspects:Research based on color firstly. According to the meaning and shape characters of traffic signs existing in our country, the paper analyses and realizes the algorithm of traffic signs segmentation based on color under different color space, and describes adaptive algorithm of color segmentation using the mean and variance of classical probability statistical characteristics under the color space of RGB. The experimental data shows that the information of color segmented through this kind of algorithm can not only ensure the accuracy and the real-time requirement of color segmentation, but also adapt to change of illumination under different conditions.Then research the algorithm of shape detection based on Fourier descriptor. Common methods are proposed for geometric analysis of object, such as Hough transform, Hu rotation invariant moment. Subsequently, shape analysis based on Fourier transform is raised using function of outline of the center distance. To apply for the condition of lacking shape characters, database of normalized Fourier descriptor template is employed in this paper. The traffic signs could be detected quickly and efficiently, especially for the condition of obscured, distorted, scaling and so on. Meanwhile, this algorithm is robust to the detection.Support vector machine (SVM) based on HOG feature is introduced subsequently. After locating the traffic signs using the above method, how to recognize the meaning of traffic signs is the key content. This paper introduces HOG feature descriptor, calculation method of statistical learning theory and SVM theory systematically. Moreover, the commonly used image library of traffic signs at home and abroad is also stated, and the optimized parameters of SVM are analyzed. It can be seen from the contrast experiment that the recognition of SVM based on HOG feature can not only guarantee the speed, efficient and integrity of feature extraction, but also ensure the accuracy of the identification of candidate regions.The last part is system realization for the algorithm of detecting and recognizing the traffic signs. Test and verify the algorithm through experiments, taking static images as test samples. Based on the statistical analysis of real-time requirement, rate of detection and recognition, it concludes that the algorithm can not only locate and identify the traffic signs accurately, but also satisfy the real-time requirement of daily. |