車牌辨識系統之建立與應用

Development and Application of a Vehicle Plate Recognition System

楊登凱(Deng-Kai Yang)

指導教授 ： 蕭耀榮

國立臺北科技大學/機電學院/車輛工程系所/碩士(2006年)

摘要

車牌辨識結合了數位影像處理與物體辨識的技術，使電腦可以自動地擷取車輛之影像，並由影像中萃取出所需要的資訊，包括車牌的位置及車牌之字元。車牌辨識可以應用的範圍非常廣泛，如大樓進出管理、電子公路收費系統、停車場收費管理、警察於贓車的查緝等。
本論文針對目前台灣的車輛車牌進行辨識，車牌辨識系統流程包含車牌影像前處理、車牌定位與字元的萃取、車牌字元辨識等三大部分。首先在車牌影像前處理方面先利用影像處理中的二值化與同色長度編碼的方法進行車牌影像的前處理與修整，將非車牌的背景與車體去除；然後在車牌定位與字元的萃取方面使用投影法在影像中找出車牌的位置，且將車牌字元的影像一一切割出來。最後在車牌字元辨識的部份，有鑑於單一辨識器其辨識率大都有一定的瓶頸，所以採用了多重辨識的觀念，集合了兩種辨識方法：樣板比對法(Template Matching)與類神經網路(Neural Network)以同步辨識的方式辨識切割出來的字元影像，而後再結合兩者的辨識結果，得到較精準的車牌號碼辨識結果。
本研究對187張自小客車影像進行測試，統計其測試結果，車牌定位成功率為94.12％、字元切割成功率為95.5%、車牌字元辨識成功率為95.83%。

關鍵字

影像辨識 ； 影像處理 ； 車牌辨識

並列摘要

Vehicle plate recognition combines the technologies of digital image processing and object recognition. It enables the computer to catch vehicle image automatically, and extracts the needed information out from the image. The information includes the position and characters of a vehicle plate. The vehicle plate recognition has numerous applications, such as building security, electric payment tollgate, management of parking lots, stolen car searching for police, and etc.
The vehicle plate recognition system includes three phases:(1)vehicle plate pre-processing;(2)vehicle plate locating and characters extracting；(3)vehicle plate character recognition. In the vehicle plate image pre-processing, we process the vehicle plate image with bi-level quantization and run-length coding. In the part of vehicle plate locating and character extracting, we locate the position of a vehicle plate and extract those characters image by using the Projection method. In the recognition part, we use an Multiple Classifier to construct main recognition system becouse every single classifier has lower character recognition ratio. In the Multiple Classifier, we adopt two recognition methods: Temple Matching and Neural Nework.These two classifiers operate in parallel to recognize the characters from image, and those two recognized characters from two classifiers are combined to increase the character recognition ratio.
Our vehicle plate recognition system examined 187 car images. The vehicle plate locating ratio is 94.12%, characters extracting ratio is 95.5%, and characters recognition ratio is 95.83%.

並列關鍵字

image recognition ； image processing ； vehicle plate recognition

參考文獻

[3] 羅華強，類神經網路入門-Matlab的應用，新竹：清蔚科技股份有限公司，2002。

[22] 仲崇實，影像處理及類神經乏晰於車輛牌照自動辨識之應用研究，碩士論文，元智大學資訊研究所，桃園，1998。

[32] 魏銪志，動態多標的車牌辨識系統之研究，碩士論文，元智大學資訊研究所，桃園，2000。

[6] Alasdair McAndrew, Introduction to Digital Image Processing with MATLAB, GL,2004.

[10] O’Neil, "A Threshold Selection Method from Gray-Level Histograms," IEEE on system, System, Man, and Cybernetics, vol.9, No.1,1979 pp.62-66.