Study On Traffic Sign Recognition Algorithm Based On Feature Fusion And Sparse Coding

Traffic sign recognition is one of the key questions in driver assistance systems andunpiloted systems. However, traffic signs which are located outdoors are easily affectedby weather and illumination. Besides, traffic sign images may be fuzzy which caused byocclusion and changes of view and movement. Therefore, large category traffic signrecognition has a broad application prospect and a great challenge. Pretreatment, feature extraction and categorized design are the key questions oftraffic sign recognition. An in-depth study about feature extraction of large categorytraffic sign recognition was conducted in this paper. The research work in this paper includes the three following aspects:(1) HOG and SIFT are the features which are adopted commonly at present. In this paper, a syncretic feature combined HOG and SIFT was proposed based on the study of HOG and SIFT. This feature was not only with the property of robustness, resulting from histogram counting in blocks like HOQ but also with the property of rotation-invariance, from main-gradient matching like SIFT. Therefore, this feature enhanced the expressive ability of the large category traffic sign recognition features and laid a basis for precise recognition.(2) An algorithm of encoding-in-two-phases was used in this paper to optimize the feature ulteriorly. LLC method was used in this algorithm and the parameters were optimized. In the choice of datum space, a multilevel datum space cascaded by K-SVD based on sample learning and DCT based on mathematical model was proposed.(3) Above feature and SVM were used to recognize traffic signs.58 categories and 2494 traffic sign images were recognized and the recognition correct rate was 99.2%. The result proved that the algorithm was effective.