Research And Application Of License Plate Recognition Based On Machine Learning And Convolution Neural Network

With the high-speedy development of social economy,the car as the previous luxury goods has become a necessity of people's daily life gradually.The vehicle management and transport operation are faced with more and more pressure and challenge with the number of automobile-owned increase with each passing day.To solve this problem,vehicle license plate automatic recognition technology(AVI)arises at the historic moment and becomes one of the most important parts of modern intelligent transportation system.License plate recognition technology was based on digital image processing,computer vision processing and relied on pattern recognition,machine learning and other means.The collected images were analyzed and processed for extraction of license plate recognition.The technology can be applied to the parking lot entrances,residential access control,highway toll management,etc.It has a positive and realistic significance for the maintenance of traffic safety and urban management,and the realization of automatic traffic management.The accuracy of current license plate recognition technology is more than 90% in an ideal experiment environment with fixed ground sensors and cameras,lighting equipment.The accuracy for commercial license plate recognition is more than 95%.The current license plate recognition was lacking in flexibility and mobility.Anti-interference and stability would be declined in some complicated cases.The further research is needed in license plate recognition field.The main samples in present paper which were used in license plate recognition were different with the images from the traditional fixed camera.The samples were from different phones,different scenarios and different angles.The open source image processing library of OpenCV was used to facilitate transplantation images.License plate was extracted by the method of combining the color and morphological of the mages first of all.The SVM method was used to positioning plate for twice.The BP neural network algorithm was used in binarization processing,denoising and tilt correction.The improved LeNet-5 was used for license plate character recognition after character segmentation.The experiment was studied with Visual Studio 2013 under Windows 7 priority.The exploration and attempt study were also been carried under Android Studio at the same time.The experiment result showed that the system has the advantages of high speed and high accuracy.It is easy to transplant to embedded devices.The identification accuracy is high in the mobile terminal and the speed needs to be further improved.