The Research And Design Of Vehicle Plate Recognition System

License Plate Recognition(LPR), a key technique of Intelligent Transportation System(ITS) and a hot research area in Image Processing and Pattern Recognition, is applied to solve the problem of automatic license plate recognition and widely used in monitoring record at traffic crossing. Nowadays, some breakthroughs occur in this area, but the effectiveness and stability of its relevant algorithms should be improved due to the complex application scene and changing illumination surroundings.The paper analyzes the theories and algorithms of LPR, presents the characteristics of LPR and emphasizes the research in LPR location, character segmentation and character recognition. Based on the 3 parts above, an application consisting of catching image in real time, downloading offline, license plate recognizing and outputting the recognition result, which is developed on Microsoft Foundation Classes(MFC) in Visual C++6.0, is designed. The work in detail is as follow:(1)Location and detection:the chapters introduce some commonly used location methods, analyze the texture features of license plate, present and put two different location methods into comparison:the first one which is comprising edge detection,edge image binaryzation, edge transition density filtering and a method based on morphological processing and connected component analysis, is based on edge transition density filtering location method. The second one is combining Haar-like feature and Adaptive Boosting(AdaBoost) to detect license plate, which puts some user-defined edge detection operator into use. The experiment shows that:the first method can realize a fast detection of license plate at a good illumination condition, however, fail to detect ones at a weak condition:rate of detection 96.40%; the second method has a better adaptation to nonuniform illumination condition, noise disturbance and spot contamination and the elapsed time meets the requirement of real time: rate of detection 98.31%. Thus, the second method is chosen to use in this system.(2)Character segment:the chapters discuss some commonly used methods in preprocessing and character segmentation and carry out this two processes. The preprocessings include:1) implement horizontal correction of license plate based on Hough Transform; 2)a novel character segmentation method based on projection curve saltation analysis is proposed; 3)a new approach based on character strokes width to judge to reverse color is present; 4)implement image enhancement based on morphology;5)a novel method based on block partition to make image binary be proposed; 6)implement an approach based on Radon Transform to detect the vertical angel and use skewing to correct the license plate. The character vertical correction:this paper presents a template matching method to segment characters based on vertical projection integral curve. The elapsed time meets the requirement of real time and the correct rate of segmentation to experiment data is 94.67%.(3)Character recognition:the chapters discuss some commonly used methods in character recognition and realize the calculation of uniform Local Binary Pattern(LBP), block partition of LBP, calculation of LBP feature histogram and using this result to recognize Chinese characters, English letters, the mixture of numbers and English letters based on Support Vector Machine. The experiment result shows that the method proposed can complete character recognition and the rates of recognition of three parts above are 95.11%,97.86%, 97.08%.(4)System design:The MFC program frame based on Visual C++ 6.0 is applied completes the design and development of LPR, which realizes the detection and recognition of the picture and video streaming capturing from industrial camera. The experiment result shows that this system can complete the recognition in real time and offline, basically achieve the goal of design. The system runs stably and obtains 88.25% of rate overall when off-line test,and 85.45% when on-line test.The paper presents and realizes some relevant methods of LPR. The elapsed time basically meets the requirement of real time and achieves a good rate of recognition and the system developed also works well.