The Design And Implementation Of License Plate Recognition System Based On Robust Illumination

As one of the core technology in intelligent transportation systems(ITS), license plate recognition(LPR) technology is growing more and more important. Although the current LPR technology has made a lot of achievements, the existing LPR systems can achieve satisfactory results only under ideal illumination conditions. The true illumination conditions vary from minute to minute, in order to make LPR technology to practical application, the research on improving the illumination robustness of LPR systems still has certain theoretical and practical significance.Based on the analysis of domestic and foreign LPR systems, according to the features of Chinese license plate, the research on feature extraction and recognition methods of license plate characters is focused to improve the recognition rate of LPR systems under complex illumination conditions, the concrete research contents include:To license plate image preprocessing, the common methods are discussed in LPR technology,including image gray-processing and binarization, enhancement, morphological-processing and edge-detection. Then the common license plate location and character segmentation technology are researched, and the suitable methods are selected, finally expounds the effect on the selection of these algorithms for character recognition.For extraction of character features, common character features and extraction methods are studied. The analysis show these methods ignore the effect of complex light, then the local binarization pattern(LBP) algorithm which has illumination robustness in face recognition field is properly modified and applied into LPR systems. Based on LBP and the elastic mesh extraction methods, this paper presents a method for extracting character features using them both.In character recognition part, the exiting algorithms are studied, get the result that network training to failure as the classification of sample spaces are big, but the convolutional neural network(CNN) can overcome above disadvantages effectively by using the improved back propagation algorithm based on gradient, so CNN is used for LPR system. Based on CNN algorithm and character image features, a CNN structure of six layers is designed. The finally test results show that the recognition rate of character image can achieve 95% under different light conditions, meeting the design requirement of illumination robustness and high recognition rate.