Study On License Plate Detection And Recognition Method Based On Visual Information

License Plate Recognition System which ranks the pivotal position among the technology of intelligent transportation system has become a popular research topic recently and has been widely developed at home and abroad. It is widely applied to deal with the problems of fled and illegal vehicles, solve the problem of highway tax loss, thus it is significant to conduct further research on the technology of license plate recognition.License plate recognition system consists of four facets as followed: positioning of the license plate, tilt correction, characters segmentation and characters recognition. These four aspects are deeply studied in this paper:In the part of license plate location, the traditional localization algorithm is improved based on the method that HSV(Hue, Saturation, Value) color model which is combined with the license plate prior knowledge. The interference of false plate area such as advertising can be effectively excluded through the method, and the plate area can be more accurately positioned when the image contains a plurality of plates.In the section of the license plate tilt correction, in order to solve the problems of low accuracy of the license plate tilt corrected through other correction algorithms, the method of feature points extraction combined with inertial axis is proposed to correct tilt license plates located in the previous step. At first, the region of the license plate located is grayed and the noise is removed with bilateral filtering. And then, the feature points of license plate area are extracted with the Harris corner detection, the inertia spindle is obtained by the feature points.At last, the tilt license plates are rotated based on the angle between the inertia spindle and horizontal or vertical direction. As a result, the correction of license plate is realized. The accuracy of the calibration is enhanced by the improved algorithm which provides a guarantee for the subsequent character segmentation and recognition accuracy.In the portion of license plate character segmentation, the method of horizontal projection combined with the structural features of characters is exploited to realize character segmentation. The license plate area is divided into two lines by the horizontal projection initially, and then individual character is segmented by structural features of characters. The performance of segmented characters in double lines license plate is proved to be effectively improved through the method.In the module of license plate character recognition, the characters are identified by the method of support vector machine which is on the basis of hybrid kernel function. Firstly, the characters features are extracted. Secondly, Chinese characters classifier,letter classifier and the mixed classifier which related to the number and letter classifier are designed and trained by structural rules of characters. Owing to Hybrid kernel function possessing the advantage of individual kernel, the performance of support vector machine and the rate of character recognition are improved.A large number of simulation experiments are performed on the platform of MATLAB in this paper. The experimental results show that the efficiency and correction rate of license plate recognition are fairly enhanced by the improved algorithm in each part of the system.