以動態投射校準技術建立之穩健車牌辨識系統

Abstract

在過去的二十年裡，車牌辨識的技術迅速的發展而且廣泛的被應用，如：智慧型運輸系統、電子收費系統、影像執法系統、…等等。然而，在實際應用上的問題卻造成系統辨識能力的下降，諸如不穩定的光線環境、模糊的車輛影像、骯髒或不完整的車牌字元。本論文提出一個結合空間遮罩及列差分分析的車牌粹取方法來改善車牌定位的成功率，其定位成功的比率達到98.9%。另外，亦針對切割時產生位移、削蝕或不完整的車牌字元設計了一個動態投射校準辨識方法，其平均辨識率更達到99.3%。若能將本論文所提出的系統以低成本硬體加以實現，將來車牌辨識技術勢必成為相關應用中普遍及重要的一環。

During the past two decades, license plate recognition technique has been increasingly developed and widely used in diverse applications such as intelligent transportation system (ITS), electronic toll-collection systems (ETCS), image enforcement system, etc. However, some practical problems impair the recognition rate, such as varying illumination outdoors, blurring image of a moving vehicle, dirty and fragmented license plate characters, and so forth. This paper proposes a novel license plate extraction method combined spatial mask method with row differential analysis procedure to improve the accuracy of license plate extraction and reaches the accuracy 98.9%. In addition, a robust character recognition method, dynamic projection warping, which is designed for shifting, paring, and fragmenting characters is also proposed in this paper and reaches the recognition rate 99.3%. If the proposed LPR system were implemented by low cost hardware, it could be very popular and practicable in the related applications.