Research And Implementation Of License Plate Recognition Algorithm On Complex Background

With the continuous improvement of China’s economic level,vehicle ownership increased year by year,more to serve our lives.The increase in vehicles to bring convenience to life at the same time,has brought great challenges on the use of roads and vehicle management.In China,the existence of license plate and vehicle there is a one-to-one relationship,so the identification of the license plate can improve the efficiency of vehicle management.License plate recognition technology this year has been one of the popular research in intelligent transportation system,but current license plate recognition products,its versatility is insufficient,that is,in complex environments such as strong light changes,tilt,fouling,noise,etc,the efficiency of identification is not high.Therefore,the research of the license plate recognition technology on the complex background has a significant market value.Based on the above background,this paper deeply studies the main aspects of the license plate recognition technology,including license plate positioning,tilt correction,character segmentation and recognition.Based on the latest technology of modern computer vision,we propose new ideas or improve some traditional algorithms.The new algorithm,after experimental verification and analysis,has achieved good results.The main contents of this paper are as follows:1.License plate location algorithm.Based on the characteristics of color information and rich edge information on the the license plate area,the edge information is detected first,then the edge information is filtered using the components of the HSI color space,after that the edge information is filled using the custom edge connection method and the morphological method,we select the candidate area of the license plate according to the geometric characteristics of the connected domain and the license plate.Accurate positioning of candidate areas,including the use of edge information for tilt correction,removing license plate border.Finally,using the SVM method,combining with the HOG and LBP features extracted from the precise positioning of the license plate,locate the real license plate area.2.Character Segmentation Algorithm.In this paper,we improved two traditional algorithms,and discussed them in detail.The character segmentation algorithm based onmulti-threshold and connected domain,the character segmentation algorithm based on character distance and binary projection.The former for the connection domain can not extract all the characters at once by one threshold.The design of multi-threshold extracts the character area multiple times which meets geometric conditions,and we proposed a non-connected Chinese character extraction method.The latter uses the characteristics of the license plate character distance to find the demarcation point between the second character and the third character,and then use the binary projection and character geometric features for character segmentation.3.Character recognition algorithms.Convolution neural network is introduced,for different characteristics of alphanumeric and characters,we design different network structure to achieve the purpose of character recognition.We improve the robustness of the algorithm,that’s mean the algorithm’s ability to deal with complex background is improved.The algorithm in this paper is implemented on the VS2010 platform,developed by programming language C ++,and we use the computer vision tool library OpenCV 1.0.