Research And Implementation Of The Moving Vehicle License Plate Recognition Algorithms

The transportation construction is an important part of national construction, and the operation on national economic level requires transportation to support. Therefore, mature electronic, network, machine vision technology are used to build intelligent transportation system (ITS). The main purpose to develop the system is to improve traffic flow, reduce traffic congestion and traffic accidents. In this paper, image processing and computer vision technology are utilized to design a set of license plate recognition systems, which is a fundemental part of the Intelligent Transportation System.The Support Vector Machine (SVM) is first proposed by Cortes and Vapnik in1995. It is a machine learning method based on VC dimension theory of the statistical learning theory and structural risk minimization principle. It aims at getting a good statistical regularity through structured risk minimization to improve the generalization ability of learning machine. SVM has been considered as one of the best available classification algorithm all the time, and it shows numerous advantages in solving small sample size, nonlinear and high dimensional pattern recognition.MQDF is based on the statistical model and the central limit theorem, performing K-L transformation on the basis of QDF. In addition, MQDF uses constant to replace the small eigenvalues to improve the calculation speed and the correct classification rate. Generally, MQDF has good robustness and high recognition accuracy, being widely used in the handwriting recognition area.License plate recognition system consists of detected the license plate, character segmentation and character recognition. Specific research contents are listed below/in the following:1. In license plate detection, traditional license plate’s detection techniques (mainly based on texture and color-based) perform low recognition rate when detecting various sizes and angels of license plates, and even license plates with flushing at night. This paper is to illustrate the cascade Adaboost machine learning algorithm that based on Haar features to detect license plate. Through this algorithm, license plate recognition rate and robustness are becoming higher, but the mistake rate accordingly rises. However, the character segmentation and character recognition that are in the following can bring feedback to eliminate false postitive candidate area.2.In the segmentation of character, this is mainly proposes a method that is based on connecting domain coordinate information and firing a straight line to horizontally correct and remove the horizontal border on the license plate. The robustness of the method is high.3. In character segmentation, this is also proposes a method which is based on the connected domain coordinate information and license plate priori knowledge to find large interval. The method shows great segmentation effect and simple implementation.4. In character recognition, this is combines the advantages of the MQDF and the SVM classification algorithm to identify the character, improving level matching strategy. The characters can be identified through wisely use of the differences among similar characters and partially use classification algorithm above.