車輛車牌辨識(license plate recognition , LPR)在現今交通監控系統中扮演相當重要的角色,其應用包含控制進出車輛數量、電子收費的過路收費、超速車子的偵測系統等等。LPR 包含三個步驟:偵測車牌、字元分割、字元辨識。而如何正確的找出車牌位置是最重要也是最基礎的工作。 目前有許多研究都以單張影格作為偵測車牌位置的來源影像,再透過一些演算法來找出車牌的區塊,車牌區塊確定後再進行車牌字元的辨識,但正確的車牌定位卻會因為照片的取樣而有所誤差,導致後續即使有再好的字元辨識演算法也是徒勞無功;所以本論文提出一個以連續影像的特性來進行車牌定位的演算法,試圖提高車牌定位的正確性,演算法共分為四個步驟(一)利用連續影格相減及區塊演算法找出移動車子區塊,(二)利用形態學原理找出每張圖片車子區塊候選車牌位置,(三)以車子路徑的斜率、位移與各車牌路經的斜率、位移關係建立車牌路徑的成本,(四)以動態規劃演算法在車牌路徑的成本找出最低的成本車牌路徑,即為最佳的車牌路徑。經過實驗驗證,利用連續影像進行車牌定位演算法在實驗影片中均可以正確的找出車牌區塊。
Vehicle License Plate Recognition (LPR) plays an extraordinary important role in nowadays transportation supervisory system. Its applications include quantity control of vehicle access, electronic toll booths, speeding vehicle detecting system, and so on. LPR involves three steps: license plate detection, character segmentation, and character recognition. The most essential and fundamental task is to correctly locate the position of the license plate. At present, many studies use single image frame as the source image to detect the location of a license plate, find the plate segment through some algorithms, then process the character recognition of the plate. However, the retrieval of the sample source image may lead to incorrectness of the plate location/position, which then causes a futile effort of those so-called good license plate character recognition algorithms. Therefore, this paper proposes an algorithm that utilizes the characteristic of continual images to process the position of a license plate to increase result accuracy. There are four steps for this algorithm: (I) use subtraction of continual frames and block algorithm to find the area of moving vehicles; (II) use morphological principle to identify candidate license plate area in each frame; (III) use the slope path of the vehicle with displacement from the plate, and the slope of the plate to establish a relationship among the plate, the displacement, and the path cost; (IV) employ the dynamic programming algorithm in the cost of license plate path to identify the lowest plates path cost, which is the best vehicle plate path. Our experiment verifies that the algorithm proposed in the paper does correctly identify vehicle license plate.