Research On License Plate Recognition Algorithm In Moving Conditions

With the rapid development of our country’s economy, the national living standard is improved continuously. Vehicles have increasingly become one of the indispensable means of transport for travel, which also makes traffic problems become prominent gradually. So how to solve the traffic problems effectively has become a hot spot in recent years. As an important part of the intelligent transportation system, the study to License plate recognition system has very important theoretical and practical significance.Based on the detailed study of the representative license plate recognition system algorithms, this paper takes the idea that detects the vehicle target in the video at firstL then recognize it. In the section of moving objects detection, a method of Gaussian Mixture Background is applied to establish the background model which can be updated in real time. To track the interest region simply can improve the operating speed of the algorithm. And by using the update method of combining Gaussian Mixture Background modelling with Inter-frame Difference, different parameter update rate is used for upcoming changes and no changes region of the imageLso that the modeling process can be completed faster.In the section of license plate location, a new algorithm of license plate location is proposed in this paperLwhich is based on Edge gradient filter combined with Gray jump reform. After the preprocessing of Gray level expand to the vehicle image, the edge detection operator is used. High frequency is reserved to the gradient of the entire image in the horizontal direction according to the feature of the license plate image edge gradient. So it can remove the interfering factors that do not belong to license plate image as possible as it can, which can acquire the candidate regions of the plate image. In the meantime, according to the license plate image color information and gray hopping feature, using algorithm of multiple determination mechanism to remove the pseudo license plate area and locate the position of the license plate from the candidate regions quickly and accurately.In the section of license plate character segmenting, this paper combines the limited threshold binarization algorithm with Otsu method to binarize the license plate image. It means that gray stretch technology is applied to limit threshold binarization algorithmLso the binarization threshold value within a certain range can be controlled. Then the Otsu method is used to obtain another binarization thresholdLand these two thresholds are used for license plate image binarization. After binarized the image, an improved vertical segmentation method is used to segment the characters. This method makes use of the length and width proportion of the license plate to verify the single character to make sure the characters can be segmented accurately.In the section of character recognition, this paper combines the template matching method with BP neural network method. Firstly, the template matching method is applied which is based on the number over the line features to extract features and recognize the character preliminarily. Then a method of BP neural network of character recognition with features is proposed, which uses learning training method with momentum factor to modify weight value. And according to the classification of license plate characters, three different sub-networks recognition is used as well. This method improves the BP network learning speed. And through the experiment, its accuracy is verified quite well.