Research On License Plate Recognition Algorithm Based On OpenCV In Complex Scenes

License plate is the unique identification of the vehicle,and its particularity and importance determine license plate recognition system to be an indispensable important component of the intelligent traffic management.However,due to the complexity of scenes and the increaseing requirement of light suitability,license plate recognition technology is also facing many problems to be solved.Applying OpenCV,this paper focuses on how to achieve license plate recognition in complex scenes.Specific work is as follows:(1)Frequently used license plate location and segmentation algorithm.The research on two location method based on color and edge detection,which is applied to compare with the method for license plate location based on MSER.Location based on color switches the sample image from RGB color space to HSV color space first,and then,license plate was screened through color component and Otsu is used in the process of binaryzation.Based on edge detection,the sample image was dealt with Gauss denoising and made it into binaryzation by putting Sobel edge detection algorithm and Otsu algorithm.These binary threshold images obtained from the two location methods above were handled with morphology processing,contour detection,rectangular search and size filtering for the coarse location of license plate area,and then,removed the false license plate area by using SVM classifier.Finally,we divided the obtained license plate area into single license plate character through the contour method.(2)License plate location and segmentation algorithm in the complex scenes.In this paper,we designed an algorithm for the coarse location of license plate area based on MSER algorithm and precisely location of license plate through SVM classifier.Coarse location algorithm are divided into three steps,MSER regions extracting;MSER filtering;MSER regions merging.Then the merging area was located precisely by the SVM classifier and removed false license plate area.Finally,we divided the obtained license area into single license plate character through the contour method.By contrast with the orientation of color and edge detection,the algorithm in the paper shows the extremely strong robustness.(3)Recognition of license plate character.In this paper,we designed the license plate recognition method based on the elastic BP neural network.We extracted the license plate characters through the coarse grid method and HOG features method at first.Then,we trained the elastic neural network.by the characteristic vector modally.Four types of recognition model were designed in this paper,which were model of Chinese characters,model of English characters,mixed model of English letters and Arabic numerals and model of easily mixed characters respectively.Finally,we used the four models obtained from the training to test the accuracy of license plate character recognition.