In recent years, with many global problems related to the environment, energy and traffic safety becoming increasingly serious, various resolutions have been raised to solve these problems by researchers from different countries, among which the study of the intelligent transportation system (shorted as ITS) is gradually becoming more and more popular in the world. Traffic signs recognition is an important aspect of the ITS, and researches and discussions have been conducted on it since1980s. However, there is no mature systems for the traffic signs recognition due to the complexity of the traffic signs and the environment. Therefore, further researches’are still needed to be carried out to get a practical and effective system for the traffic signs recognition.According to the location of the traffic signs, they are divided into air traffic signs and road traffic signs. As follows, the researches on the three typical road traffic signs (go straight, go straight and turn right and turn left or right) are carried out in this paper:1. In order to solve the image distortion problems caused by the camera angle, an implicit model of the camera is built and a method for getting the control points quickly is used to conduct the image scene reconstruction;2. Because of the complex environments of the road traffic signs, there are no ideal effects for the whole image segmentation. Therefore, a method based on the regions of interest for the image segmentation is utilized in this paper. The regions of interest of the road traffic signs are established by using the Hough transformation method to detect the traffic lane. Then the target image can be obtained by conducting the one-dimensional entropy segmentation, the image filtering and the Canny edge detection on the regions of interest;3. In order to build the image feature vector which can characterize the image effectively, the traditional Hu invariant moments are improved to get the better ones which possess the performance of translation, rotation and scaling invariability in the discrete case. We should extract the features of the improved Hu invariant moments and those of the affine invariant moments, and conduct the comparison and analysis on both. The features that possess good aggregation and separability will be chosen to establish the image feature vector;4. To achieve good recognition results and promotion performance of the road traffic signs classifier, a1-a-r multi-class classifier based on the support vector machines (SVM) is used. The type of the kernel function and the relevant parameters are determined for the classifier through experiments; 5. The methods used in this paper are verified through experiments. |