Research And Implementation Of License Plate Character Recognition Technology

With the rapid development of the economy and the steady improvement of people's income,private vehicles have become the primary choice for people to adapt to the fast pace of travel.However,the construction of transportation infrastructure in China does not keep pace with the automobile growth,which lead to a series of social problems,such as traffic congestion,environmental pollution,parking difficulty,etc.And It restrict the economy development seriously.Therefore,more and more scholars have joined the team in research of traffic problems.Among them,the smart transportation technology is an important method to maintain the sustainable development.The license plate recognition technology is one of the important research topics in the smart transportation system.The license plate character recognition is also an important supportive point of the license plate recognition system.According to the research results and experiment from pioneer scholars,this paper proposes a new improved algorithm for several key technology of license plate character recognition.Firstly,image preprocessing is performed by using an improved image enhancement algorithm based on top hat transform.After the brightness is balanced,the license plate image can eliminate the noise generated by uneven illumination changes during thresholding process,and effectively solve the noise problem of license plate image background.It lays the foundation for the subsequent character recognition of the license plate.Secondly,by analyzing the commonly used template matching functions,a new license plate matching function based on global information,the global coincidence function,is proposed.The experiment shows that the global coincidence function can significantly improve the efficiency in recognizing the license plate character,This is more advanced than the traditional template matching function.Thirdly,based on the specific of license plate character image,the character feature extraction algorithm and the feature classifier are optimized respectively.The experimental results show that the recognition rate of the optimized license plate recognition algorithm is 96.54% and the average recognition time is 0.384 s.Finally,on the basis of the traditional LeNet-5 convolution neural network structure,this paper proposes Multi-Simplified Convolution Neural Network,M-SCNN.The use of MSCNN network structure can greatly shorten the time for character recognition,the average recognition time is 0.003 s with 96.23% of recognition rate.