Research And Implementation Of Traffic Sign Recognition Technology Based On Color Information And SVM

With the development of the socio-economic and urbanization, motor vehicles has been rapidly increased and resulted in increased traffic congestion and frequent traffic accidents. Public transport and traffic efficiency have become increasingly prominent. Intelligent transportation system (ITS) is proposed to solve the increasingly serious road traffic conditions. As an important part in ITS, traffic signs recognition (TSR) has great practical significance and application value.TSR mainly includes two stages, the first is traffic sign detection and the second is traffic sign recognition. The algorithms of the two stages were studied and implemented in this paper. The first stage was the traffic sign detection. In this stage the color threshold segmentation algorithm after image preprocessing was adopted for segmenting the image preliminary. After segmenting many interference regions were reserved. The binary-area-threshold algorithm was adopted for filtering the interference regions, and the algorithm was improved as follows: First, the binarization method of original algorithm was based on single-threshold. The double-threshold binarization method based on Otsu algorithm was adopted for increasing the image contrast. And then the area threshold was selected. An adaptive threshold combined with the empirical threshold selection method was adopted. The second stage was the traffic sign recognition, and the two grades classification and recognition SVM-based was adopted. Edge orientation histogram was adopted for describing the external contour of traffic sign in the first classification. And the sensitive features selection was adopted for increasing the classification accuracy rate. The basic concept of the sensitive features selection was to make the within relative error minimum and the between relative error maximum. According to the effect on the classification of the sensitive features, the contour features were weighted after selecting the sensitive features. And then the traffic signs were classified. Finally, the traffic signs were divided into three categories:round, triangle and rectangle. Hu invariant moments were adopted for describing the regional feature of traffic sign in the secondary classification and made a further recognition of the traffic signs. Finally, standard images and text commentaries were showed.The test results show that the traffic sign detection rate is increased by the improved binary-area-threshold algorithm and the interference regions are filtered in maximum. In the first classification, the classification accuracy rate is great improved by the method of the sensitive features selection. Finally, test has been done on robustness and accuracy rates under different illumination conditions and block degrees. The result indicates that the algorithm has better accuracy rate and the robustness is also better under different situations.