Research On License Plate Recognition Algorithm In Complex Environment

In recent years,with the improvement of economic level,vehicles are increasingly popular among the people of our country.This makes traffic pressure of road increases day by day.Those problems like traffic congestion,traffic pollution and traffic accidents are put on the agenda urgently.At this time,Intelligent Transportation technology came into being.As is known to all,vehicle is the most important unit of the transportation system,and the license plate is the only identity of a vehicle,we can accurately record its traffic behavior by the plate,so License Plate Recognition(LPR)is one of the most significant foundations in intelligent transportation field.Through reviewing many documents we can find that there still exists many problems of LPR based on image identification in the complex environment.So this dissertation will make a deeper study of LPR in the aspect of complex environment.In the condition of complex environment,license plate location algorithm will be interfered in these cases like the external environment or there are multiple plates in one picture,or the plate tilts.Therefore,this dissertation will propose a new algorithm which combine the improved vertical edge detection and color location to search for the plate roughly.We add the parts that repair broken part and do sobel search several times to the algorithm of edge detection.This will avoid the cases like incomplete and inaccurate location.Combining two algorithm is to filter similar regions and reduce the dependence of single algorithm on their applicable conditions.Thus,the hardiness of our algorithm will be increased greatly.At the end,we will extract the feature of horizontal and vertical histogram and use SVM to filter of non-plate part.So the accuracy and hardiness of the algorithm will meet the needs in practical applications.In the part of character segmentation,based on the fact that traditional algorithms can not split the Chinese characters precisely because of the bad connectivity of the characters.Therefore,this dissertation will propose a new algorithm which will first find the character that marks the city,then guess the location and size of the Chinese character by special algorithm.Through these,our accuracy of character segmentation algorithms will be superior to traditional methods.In the condition of complex environment,character recognition algorithm will be interfered by the abnormal plates.Thus,we need to extract features of numbers and English characters separated to Chinese characters,so that we can analysis those features narrowly.At the same time,we use adopt machine learning algorithm to train plurality of samples.So that the algorithm will have the ability of self-learning and adaptability.To numbers and English characters,we extract their outline features,to Chinese characters,we extract their synthesized Hog features.At the end,we adopt SVM to train and test those samples.The data of the experiment indicates that our algorithm has achieved a more satisfactory result.