Abstract: In order to effectively improve the accuracy of the traffic sign classification, a new method was proposed through fusing the global and local features. First, local binary pattern(LBP)feature was extracted which could describe the internal texture information of traffic sign image, and then histogram of oriented gradient(HOG)feature which could represent shape information and global gist feature with description of the rough outline of the image information were extracted, and then linear combination was used to achieve feature complementary. The principal component analysis(PCA)was used for data dimensionality reduction. Final traffic sign training and classification was carried out using support vector machine(SVM)classifier. The experiments showed that with respect to a single feature extraction classification of traffic signs, the algorithm based on multi-featured fusion achieveed higher classification accuracy, but also met real-time requirements.

Key words: Gist feature, feature fusion, local binary pattern, traffic sign classification, HOG feature