Research Of Vehicle License Plate Recognition Based On Variable Universe Neural Network

With the development of national economy, the amount of automobiles is increasing substantially year by year, so it is a major problem for the management of road traffic. Nowadays, the appearance of Intelligent Traffic System (ITC) has relieved the traffic pressure effectively. At the same time, it has taken place of traditional traffic control system gradually. License Plate Recognition (LPR) is widely used in ITC, and research on LPR will promote the development of ITC with a practical value. The main contents of this paper are shown as follows.Firstly, the vehicle license plate image is the objective of LPR. Because of external disturbance, it can't be at its best, so it is necessary to carry on preliminary treatment, including image pretreatment and license plate location. After processing, we will get a binary image. On the basis, the paper presents an algorithm based on horizontal projection to locate the horizontal position, and the vertical position is determined by mathematical morphology and vertical projection, which is convenient for character segmentation.Secondly, we do some preparation as well before character segmentation. At first, we need to set the image to a unified form. The paper selects the black background and white character as the standard. If not, do the anti-color processing. If the position of the license plate has tilted, we can use Hough transformation to correct the two lines in the horizontal direction. At last, according to the characteristics of the size of license plates, we find the location of characters of the license plate in order to separate them.Lastly, there are a lot of methods for LPR. In this paper, the neural network is used to identify the license plate. The paper focuses on the analysis of the principle of BP neural network, and presents the variable universe neural network applying the theory of variable universe to adjust learning rate dynamically. Although the neural network has the function of feature extraction, the paper presents a novel algorithm based on wavelet moment for feature extraction in order to accelerate the speed of LPR. Also, according to the position of characters, three corresponding neural networks are designed throughout the whole course, and the final result is made up of three results. The experimental data show that the variable universe neural network model is amazing at the results.