本論文提出一種能夠使用在可攜式裝置上,即時擷取台灣的機車車牌影像,並能過濾複雜背景且快速定位多個車牌的方法,解決以往在車牌定位常因為複雜背景造成準確率不足與運算過多無法即時處理的問題。本文提出的為兩階段的車牌定位方法,第一階段以色彩特徵與Sobel邊緣偵測為基礎,用型態學方式找出影像中車牌的大致區域,第二階段再用改良式 Hough Transform找出影像中精確的車牌區域,並以區域是否符合車牌的大小、顏色、長寬比例與邊緣數量條件篩選出可能的車牌位置。 由實驗與結果的分析來看,本篇論文之演算流程的確有效地降低複雜度並保有不錯的執行速度,且於不同的距離與角度的測試下,保有了高定位成功率的優點,可謂是相當實用之系統。
Locating license plates from images is one of the interesting topics in pattern recognition and has many applications. Due to the complicated background in the images, it is not easy to locate the plates in real time and precisely. In particular, when the computing power of devices is limited, such a task becomes more difficult. This thesis considers the mobile devices, like netbooks, of which the resources are limited and proposes an effective approach to locating the license plates from images having complicated background. We use the license plates in Taiwan as examples. There are two stages in the proposed approach. In the first stage, the candidates are found according to Morphology based on the characteristics of colors and Sobel edge detection. In the second phase, the refinement on the candidate set is done using the adapted Hough transformation and the characteristics of the plates. The proposed approach is validated through experiments. According to the experimental results, the proposed approach can locate the plates more precisely and reduce the computing complexity; thus accelerate the computation.