Research Of License Plate Recognition System Based On Adaboost And Support Vector Machine

Collecting volume of vehicles increased rapidly with the development of China’s Economy and the progress of modernization, which brings a huge impact to vehicle management. The management quality and traffic conditions of roads can be improved fundamentally by exploiting new technology to transform backward traffic management and transport system. Since license plate is the only reliable identification of vehicle, license plate recognition has been widely applied in the occasions such as toll management and control of highway, checking and surveillance of road vehicles, vehicles parking management, vehicle access control, etc. Efficient License Plate Recognition System (LPRS) can significantly improve the road network capacity and service quality and reduce labor. It has become an important component of modern Intelligent Transportation System(ITS).License Plate Recognition System typically has three core parts, including license plate area locating, license plate character segmentation, and license plate character recognition. This paper discusses overall design of the system based on extensive research of literature and existing research results widely and in-depth, doing intensive research on each key technology of three parts and achieving satisfied recognition effect on Chinese license plate in all cases such as different scenes, different lighting conditions and different size of vehicles. The main research contents and innovations include the following three parts:The area of license plate is co-located through jointing detection method which combining AdaBoost algorithm and projection method to promote the accuracy rate of license plate locating successfully in license plate positioning stage. Only using AdaBoost algorithm to locate license plate is lack of robustness, and it needs to find millions of negative samples to achieve better detecting effect. The classifier can be trained easily through few thousand negative samples but the license plate will be located incomplete or too wide by using easy trained classifier, then lead the locating accuracy rate declining seriously. In order to overcome these difficulties this paper uses AdaBoost algorithm to detect the row of license plate in the image firstly, do some pretreatments such as graying, edge detection and binary to the image of the row of license plate and then locate the area of license plate precisely through vertical projection in the handled image of the row of license plate. The detection accuracy can reach above95percent in the real traffic scene at different times through joint detection method.Since not every character can obtain favorable segmenting effect through single threshold binary algorithm, a multi-threshold binary segmenting method is proposed in license plate character cutting stage. The color of characters is reserved if it is black otherwise it is remained. Binary the graying license plate image by the threshold from0to255and hold a set of valid thresholds, then the characters in the binary images achieved through the valid thresholds are cut out by projection method and the same or similar character areas are merged to obtain seven complete license plate. The accuracy of license plate segmentation algorithm can reach99percent in data set of this paper.The cut license plate characters are recognized through a method which combines Integration Feature Extraction and Support Vector Machine in character recognizing stage. Some pretreatments such as extracting effective area of image and refinement are done to cut license plate characters firstly. Since individual character feature cannot achieve good recognizing results and the recognition of Chinese characters is ignored, integration features of license plate characters is extracted and support vector machine exploits extracted features to build two kinds of classifier networks, one is applied to match and recognize Chinese characters, while the other for letter and digit. The recognizing rate of integrate algorithm which combines IFE and SVM can reach96percent while the robustness is favorable.Above all, license plate recognition system of this paper has high accuracy and the total recognizing speed is300ms. Furthermore, it can adjust to complex environment, so it has a certain reference value for the practical application. All algorithms in this thesis are exploited using C and C++language based on OpenCV and the developing tool is Microsoft Visual Studio2010.