Research On Chinese License Plate Recognition In Complex Scenes

With the accelerating pace of people's lives,the popularity of the car is getting higher and higher,but with a substantial increase in vehicles,road backlog,urban congestion,management complexity and other issues have become more prominent.Intelligent transportation system can allocate resources efficiently,improve the traffic capacity and traffic management efficiency,and thus ease the pressure of urban traffic jam.The license plate recognition technology as the core technology of the intelligent transportation system,is a cross fusion of multiple fields of image processing,computer vision,pattern recognition and machine learning,is widely used in traffic monitoring,parking lot management,intelligent charging,electronic police etc..In fact,the license plate recognition technology can be embedded into the portable charging machine,recorder,and even mobile terminals such as mobile phone,and make intelligent transportation more convenient,more close to life,but because the camera does not remain fixed,and the position of license plate in the image is random,which has put forward higher requirements on the technology of license plate recognition.Therefore,there is very important practical significance to the study of Chinese license plate recognition system in complex scenes.This paper focuses on the positioning,correction,segmentation and recognition of the Chinese license plate recognition in complex scenes,and puts forward some effective measures to improve it.Aiming at the problem of the different light and complex background,the dissertation improves the license plate location method base on convolution neural network and proposes the location method based on multiple color spaces.The method base on convolution neural network achieve high-precision positioning through continuous learning and training,the location method based on multiple color spaces improve the accuracy and efficiency of vehicle license plate positioning by extracting a variety of color information.Through extracting different color information,it can improve the accuracy and efficiency of license plate location.According to the problem of the angle of the shooting,the instability of the equipment may result in license plate tilt,the correction method based on the longest straight line is proposed.By correctinglicense plates before character segmentation,it can reduce the difficulty of character recognition.In view of the problem of the road bumps,the background noise and so on possibly create the character fuzziness,the border,this dissertation proposes the simple segmentation method based on the projection image.By setting threshold to segment characters,it improves the accuracy and practicability of character segmentation.Because it is difficult to identify the rotation and deformation character,this thesis puts forward a character recognition method based on length feature,which represents the contour information of the character.By extracting length feature,it can simplify the character recognition process and improves the rate of character recognition.