| The 21 st century is the century of intelligent road traffic. Intelligent transportation system, as a real-time, efficient, integrated transport and management system, has been widely used in intelligent scheduling system of urban traffic, intelligent scheduling system of highway, automatic control system of a motor vehicle. License plate recognition system, as an important part in intelligent transportation system, has played an important role in the management and control of modern traffic and also made the study of the license plate recognition technology become a hot issue in the current. The process of license plate recognition can be divided into several stages, mainly including image preprocessing, license plate location, character segmentation and character recognition. The key technologies of image preprocessing, license plate location and character recognition have been researched carefully in this paper and the primary works are as follows:(1) Image gray-scale, image enhancement, image filtering, image edge detection and binary have been studied in the stage of image preprocessing. The series of technical process have been used to improve the quality of image, remove interfering factors as well as highlight useful information on the license plate, thus facilitating the subsequent recognition operations.(2) In the stage of plate location, a quantitative description model of blue-white colored pixel is built on basis of RGB and HSV dual representation by improving the definition of blue-white colored pixel in searching colored point pair, and then a plate localization method is presented by combining searching colored point pair with regional statistical features according to newly proposed model. Firstly, the connected regions are retrieved by improved search of colored point pair and morphological processing. Secondly, connected region is clipped by exploiting corner and texture feature to complete the iterative searching match in order to acquire coarse region which contains car license. Thirdly, plate region is filtered by means of vertical/horizontal projection, Hough line detection and color retrieving. Experimental results show that this method has a high recognition rate and strong robustness for the plate images under different illumination, background, bright colors.(3) In the recognition stage of character, the way of combining statistical and structural characteristics based on the coarse gridding has been utilized to extract the features of the normalized characters and the SVM based on RBF kernel function has been used to construct models of Chinese characters, letters and alphanumeric characters for completing character recognition. In order to improve the efficiency of character recognition based on SVM, the improved genetic algorithm has been adopted to optimize the parameters of SVM. At first, the chaotic motion is merged into selecting option, then the adaptive crossover probability is introduced and the heterogeneous crossover strategy based on individual similarity is proposed, finally the adaptive mutation probability is used to replace the stable mutation probability thereby enhancing the searching efficiency of the algorithm and avoiding it falling into convergence early. Experimental results show that the SVM models based on improved genetic algorithm have higher recognition rate compared with other parameter optimization methods such as gridding method and the traditional genetic algorithm and faster training speed compared with gridding method. At the same time, the training time and the performance of the models based on the method proposed in the paper have been improved compared with the ones based on the BP neural network, and the license plate recognition system has been implemented to verify the validity and reliability of the methods proposed in this paper at the last. |
PDF Full Text Request