Realization And Optimization Of Intelligent License Plate Recognition System

With the social and economic development in recent years,the traditional vehicle management fails to meet the demand of the fast-growing number of vehicles while the intelligent transportation system(ITS)begins to receive attention.License plate recognition(LPR),as one of the core technologies of ITS,has been widely discussed.The LPR algorithm usually includes four subsections: plate localization,plate orientation,character segmentation and character recognition.However,the traditional license plate recognition system has poor recognition performance under the complex background and oblique shooting angle.Thus,This paper accomplishes a complete algorithm for LPR and makes some optimization,finally builds a complete LPR system available on hardware.The experimental results show that the LPR algorithm has superior performance.This paper adopts color localization,edge detection and Support Vector Machine(SVM)to carry out plate localization.The traditional license plate localization algorithms have their own limitations: localization methods based on gray images are ineffective in the environment with rich texture and edge features;localization methods based on color images are ineffective in the background with complex color.In order to improve the robustness of the license plate localization algorithm,this paper combines several license plate localization methods.First,detecting the possible license plate regions with color localization and edge detection techniques,and then correcting them by tilt correction and skew correction,finally using SVM classifier to select out the right plate.The experimental results show that the localization method in this paper can effectively locate the license plates from the images,and 99% can be successfully located.An improved method for license plate character segmentation based on connected domain labeling is realized.Since most of the Chinese characters in the license plates are made up of several connected domains,the traditional character segmentation methods based on connected domain labeling can only segment the letters and numbers.The paper improved the function by repositioning Chinese characters.The experimental results show that the character segmentation method in this paper can effectively segment the characters from the license plate images,and the segmentation accuracy is 99.5%.The paper uses an improved BP-neural network-based method to enable character recognition.In the traditional BP-neural network-based character recognition methods,all license plate characters are extracted with the same features.with which the recognition rate of Chinese characters is about 90% because their structure is much more complex than letters and numbers,In order to improve the recognition rate of Chinese characters,this paper makes a more comprehensive feature extraction of Chinese characters: training the recognition model by using the histogram feature of Chinese character image and the value of each pixel.Totally 600 Chinesecharacter images and 2,500 letter and number images were tested with trained models,among which the recognition rate of Chinese characters reaches 94.7% and that of letter and number reaches 98.6%.Utilizing Raspberry Pi and the Raspberry Pi camera,the paper establishes an integral LPR system through transplanting the LPR program into the Raspberry Pi.Once activating the Raspberry Pi camera to capture photos,the algorithm enables plate recognition of these photos and then activates the GPIO of Raspberry Pi to respond based on the comparison between the recognized data and the data in the license plate database established by MySQL.