- 學位論文
適用於臺灣各種車輛之車牌辨識系統
A Universal Plate Recognition System for All Vehicle Types in Taiwan
摘要
隨著經濟的快速發展，交通工具成為人類生活中重要的一環，其中各式汽車的數量激增，衍生出相當多的交通和車輛管理的問題。自動車牌辨識對疏解此類問題可扮演重要的角色。 自動車牌辨識系統的研發已有二十幾年以上的歷史，但早期受限於硬體設備的不足，如CPU及RAM等處理速度不夠快等，導致發展受影響，然而隨著電腦技術的提升，這些限制已不存在，因而吸引更多人投入這個領域的研發。在過去的研究中，少有針對各種不同種類的車輛(如轎車、小貨車、遊覽車、聯結車等)探討其車牌辨識的方法。因此本文提出一套適用於臺灣各種車輛之車牌辨識系統。 在該系統中，首先進行邊緣偵測、彩色資訊轉換以及群組化等處理，以從一張影像中找出可能車牌的區域，再利用車牌的幾何特性，達到初步車牌定位的目的。框取的車牌經由偵測轉換後，不論是何種顏色的車牌，都統一轉換為白底黑字，再微調四方邊界，去除發牌地和螺絲孔區，找出真正感興趣的車牌區域。利用投影法，精確抓出字元號碼的四方邊界，再運用連通元件標示法來完成字串切割，切割出獨立的字元。最後將經過正規化後的字元以樣板比對法進行字元辨識，達成車牌辨識的目的。 為考慮將來有多方面的應用，而針對影像中偵測任意方位、位置的車牌取像，最後實驗結果顯示，車牌定位成功率為 91.73%，車牌統一轉換為白底黑字的成功率為97.17% ，車牌辨識率為91.27%。本研究所提的系統確實可適用於對臺灣各種車輛之車牌的辨識，有別於一般只能針對特定種類車輛的系統。
並列摘要
With the fast development of the economy, transportation has become an important element in modern human life. Among the various types of transportation, the increasing number of automobiles causes a lot of problems in traffics and vehicle management. The automatic vehicle license plate recognition system can play an important role to diminish these problems. Automatic vehicle license plate recognition systems have been designed for more than 20 years, but the development of these systems was limited by unsatisfactory constraints in the hardware computation capability, such as the slow processing speed of CPUs and RAMs. As the improvement of the computation technology, those constraints no longer exist and hence more and more people are attracted to participate in the research and development in this field. Very few past studies were focusing a single plate recognition method that concurrently supports different types of vehicles, such as cars, trucks, buses, and contain trailer. Therefore, this thesis proposes a universal plate recognition system for all vechicle types in Taiwan. In this system, special processes including edge detection, color information transform, grouping, etc. are initially carried out in order to locate a possible vehicle license plate in the image. Geometry characteristics of the licence plate are then used to achieve the objective of preliminary plate positioning. The cropped plates are converted into black characters and white background no matter which original colors of the plates are. Furthermore, the image is fine-tuned by chopping out the plate originating information and the screw holes to find out the interested vehicle license plate region. Projection is used to accurately identify the four borders of characters and connected component labeling is used to segment the plate string into discrete characters. Finally, normalized characters are recognized using template matching to achieve the objective of plate recognition. In order to be applied for multiple applications in the future, images are intended to be captured in various conditions including different viewing angles and distances. The experimental results show that the system achieves 91.73% success rate in plate positioning, 97.17% success rate in plate color conversion to black characters and white background, 91.27% success rate in plate recognition. The system proposed in this thesis, which has been proved to be capable for recognizing plates of all vehicle types in Taiwan, is different from other common systems that can only work with a specific type of vehicle.