為了得到監視器畫面中車牌的資訊，我們提出一個系統來對車輛的車牌部份做偵測與定位，道路監控的畫面中，車子不會像停車場這種場合可以讓我們取得較好的畫面，所以面臨旋轉車牌影像的偵測是必須考量的。我們先利用黑白相間的Haar-like特徵訓練 cascaded分類器(Classifier)，據以找出畫面中的車牌，並且框出車牌的位置。首先將運動物件以背景相減(Background Subtraction)與連通區塊(Connected Component Labeling)抽取出來，並對其做霍氏轉換(Hough Transform)，找出旋轉角度，並將圖片校正。然後再以訓練好的分類器對校正過後的圖片作車牌偵測。我們提出的方法，以 314/394張車牌正 /負樣本做訓練，以 upright和skewed Haar-like特徵訓練出upright和skewed車牌分類器，再以860張影像做測試，實驗結果的偵測率達 87.4%，驗證本系統之可行性。對於水平旋轉所造成一般車牌分類器偵測不到車牌的情況，會有不錯的改善。

In this paper, we developed a system to automatically detect and locate license plates of vehicles. The vehicle in the real world traffic surveillance would have rotated license plate in the captured image. Due to the black-white patterns in the license plate image, cascaded classifiers using Haar-like features could be trained to detect license plate. If there is any vehicle extracted by background subtraction and connected component labeling, we can do Hough transform on the extracted vehicles for rotation correction. License plate of the corrected image could be then detected by the trained cascaded classifier. Our proposed system was trained with 314/394 positive/negative samples and tested with 860 vehicle images. The detection rate was 87.4% which demonstrated the feasibility of the proposed system.