Research On Two - Stage Traffic Sign Recognition Method

Traffic sign recognition is an important part of intelligent transportation system(ITS). After decades of research of foreign and domestic scholars, the theory and practice in the field of traffic sign recognition has gradually formed, and made many breakthrough progress. Because the natural scene is complex and there are many different kinds of traffic signs, traffic sign recognition is still a challenging problem. Considering traffic signs has many categories, this paper mainly study the two-phase method for traffic sign recognition.Firstly, according to the function of traffic signs, this paper design a two-phase method for traffic sign recognition based on PCA-LDA. Traffic signs have similar pattern design in general, if it has similar function. Therefore, traffic signs will be divided into four similar subclasses, namely the speed-limiting class, warning class, directive class and no-rules class. First, using the method of combining PCA and LDA to do a rapidly classification, namely determine the subclass. And then we apply the spare representation method to do the final classification within the subclass, namely get the specific category. Experiments show that the two-phase method for traffic sign recognition based on PCA-LDA is superior to the combination of PCA and LDA method and the two-phase sparse representation method based on local dictionary.Then we presort the traffic signs as five subclasses according to the shape and color feature of the traffic signs, namely red circle class, red equilateral triangle class, blue circle class, white circle class and no-rules class. Then in the first phase, we use the combination of HOG feature and SVM classifier to determine the subclass. In the second phase we propose a method for extracting the core region of traffic signs and use the sparse representation method for the final classification within the subclass. Through the experiment show that the division scheme of similar subclass based on the shape and color feature of the traffic signs is superior to the division scheme based on the function of traffic signs, if we use the same feature extraction method. The method based on HOG feature is superior to that of based on PCA-LDA on the basis of using the same division scheme and can also achieve better effect for the final recognition.Finally, we propose a two-phase method for traffic sign recognition based on multi-feature fusion. In the first phase, we can respectively determine the subclass by color feature and edge feature. However, single feature is difficult to fully describe the traffic signs, so we use the fusion of color histogram and HOG feature to determine the subclass. Then In the second phase we use the fusion of LBP feature and HOG feature to perform the final classification within the subclass. Experiments show that the method of multi-feature fusion can get higher recognition rate and better robustness, its recognition rate can reach 96.9%..