Research And Implementation Of License Plate Character Recognition

License plate recognition (LPR) is an important research subject in the area of intelligent transportation system. And it is widely used in the parking lot management, electronic toll collection, urban traffic surveillance and so forth. After years of joint efforts made by domestic and foreign researchers, this technology has been developed rapidly and achieved a lot of significant results. However, in some complex applications, LPR is far from perfect due to the influence of various types of vehicles, different weather condition, variable illuminance, fast moving speed, shape deformation and so on. This thesis will research and implement the license plate character recognition technology by analyzing the difficulty of existing algorithms. The main works in this paper are as follows:1. A new algorithm for tilt correction based on center of gravity (CG) of license plate will be proposed, which figures out the angle by seeking the CG of image. It is simpler and more efficient than Hough transformation. So it can be fit for real-time application.2. By researching on the existing license plate character segmentation, a segmentation algorithm based on license plate prior knowledge and the vertical projection will be proposed. This algorithm will judge whether the segmentation is correct by combining the prior width of character and the pixels of projection. It will avoid the influence of chaotic pixels, solve the problems of character adhesion and segmentation error. So it can improve the accuracy of character segmentation efficiently.3. Prominence of probability will be used as the evaluation criteria to analyze the character feature and select the optimal feature set. Feature extraction will be processed based on the combination of SVD and LDA. It will reduce the dimension of combined feature vector. And it will be more suitable for character recognition because of better distinguishing performance among different categories by eigenvector projection transformation.4. A new license plate character recognition method based on the improved AdaBoost will be proposed, which classifier is support vector machine whose kernel function is Gaussian RBF. It can be adaptively adjust the parameters according to the classification of the training results in the iterative process. Compared to traditional AdaBoost algorithm, it has improvements on selecting the model and generalization, and avoids the premature stop of algorithm. Therefore it is more suitable for the recognition of license plate character.