Traffic signs from natural scenes are received from the video on board and to be put to computers. Then through the image processing using the developed algorithms in computer, traffic signs can be recognized at real time. But the study of recognition of traffic signs from natural scenes is more difficult and with more challenge than the study in non-natural scenes. This dissertation combines the practical situation in China, Image Library for traffic signs, Image pre-processing algorithm for traffic signs, Image feature extraction of traffic signs, classification of traffic signs and Criterion for traffic signs have been developed in this dissertation. The content and the main results achieved are as follows:(1)Because the specific form of traffic signs in different country are not identical. And domestic traffic sign recognition starts late. Therefore, there are not yet publicly available database of traffic signs in china. Image Library for traffic signs in two ways. Firstly, transform the standard traffic sign image by preprocessing transformation (such as rotation, distortion, adding noise) to expand the number of traffic signs, improving the robustness of recognition; Secondly, capture the traffic signs through different angles and distance in the real scene, and then make up traffic sign image database by cutting out traffic sign from the real image with detection algorithm in our laboratory.(2)For the classification of traffic signs and the traffic signs need to identify, summed up a hierarchical coarse-to-fine classification strategy and apply it to identify the traffic signs, this improve the classification speed and robustness.(3)Gabor feature extraction and two-dimensional principal component analysis applied to traffic sign recognition, and base on lighting control preprocessing, Gabor feature extraction, two-dimensional principal component analysis and classification(include template matching, BP neural network and support vector machines), propose four kinds of recognition methods of traffic signs. Then the different image library was compared using the same methods. It is found image database through the light control can improve the identify rate. And under the same image database, various traffic sign recognition methods conducted in-depth comparison of experiment about the recognition rate and run time.(4)Some researches have been done in reducing the error recognition rate of traffic sign recognition and increasing differentiation degree of the features of the correct detection and error detection. According to the discrimination between the signs detected and the wrong detection images is not high, the unit of the vector is introduced. Gabor feature extraction gain texture features of images to increase the discrimination between the signs detected and the wrong detection images, and set the judgment formula. The experimental results show that the method achieved a very good effect in terms of the final determine of traffic sign recognition. |