Intelligent Video License Plate Recognition Algorithm Based On Hisilicon Platform

Intelligent video technology, as a trend of future surveillance, is not only a digital imagesurveillance analyzing technology, but also it is a representative of more high-end application ofdigital video network surveillance, and hence is considered to be the 4th generation of videosurveillance technology in the industry, one of the most cutting-edge mode of application in thefield.Intelligent video license plate recognition technology based on Hisilicon platform sets thevehicle behavior as its object of analysis. Through the automatic recognition of vehicle behaviorand the capture of the vehicle’s images, the technology collects accurate and effective raw data ofvehicle’s behavior management for the back-end platform application system, hence thetechnology helps the effective regulation of vehicle order and the automatic obtaining of illegalbehavior’s evidence. Therefore the technology has a good application potential and very broadapplication space. In recent years, the study on intelligent license plate recognition has become ahot issue.The thesis first analyzes the research status concerning intelligent video surveillancealgorithm, and emphasizes the intelligent vehicle plate recognition system as its research areawhile it also argues comprehensively the 4 key modules in the system: vehicle imagepreprocessing, license plate positioning, license plate character segmentation and license platecharacter recognition. As well, the thesis designs and realizes the license plate recognition systemon Huawei Hi3515 platform. Based on the study on the existing research results, the thesisanatomizes the emphases and difficulties in the design of the 4 modules for the enhancement ofthe system’s robustness and environmental adaptation. The thesis also combines the theoriesconcerning digital image processing with practical application and proposes correspondingsolution in accord with actual situation.In the image preprocessing module, the thesis studies the vehicle image’s enhancementtechnology, adopting the preprocessing methods as colored image graying, gray stretch, histogramequalization, median filtering, Roberts edge detection and binarization of iterative threshold etc.In the actual captured vehicle images, there’s high frequency of license plate region gray variationand also there are huge gray contrast between the license plate character and image background.Based on these traits, in the section of license plate region positioning, the thesis proposes awindow scanning license plate positioning algorithm based on edge gradient and it further verifiesthe positioned plate region on the basis of projection feature and rivet feature and hence to a certain extent it improves the accuracy of license plate region positioning, providing a solidfoundation for the consequent license plate character segmentation and recognition. In the licenseplate character segmentation module, based on the preprocessing of image, the thesis adopts thecharacter-template-based segmentation algorithm in line with the features of license platecharacter, and processes character’s overlapping segmentation by utilizing the feature thatcharacter envelope and vertical projection of character binary image show trough simultaneously,therefore the thesis effectively overcomes the overlapping of character and the noise interferenceof license plate’s 4 frames and consequently improves the segmentation quality, endowing thealgorithm excellent robustness. In the module of license plate character recognition, the thesiscompares the existing research results and employs the BP-based neural network technology, andhence improves the accuracy of character recognition and enhances the stability of the system.The algorithm used by this thesis has been realized and successfully operated. As shown inthe experimental results, the entire system has a simple algorithm with low complexity, greatrobustness and environmental suitability, and it will present practical value in the areas of EPCsystem network security, intelligent traffic and security surveillance.