街道圖中道路及格線之抽取

Abstract

本研究將自動擷取地圖中道路及格線的資訊，以便于建成有用的資料庫系 統，供查詢之用。第一部分，我們使用Hough 轉換偵測地圖中格線的位置 。其中包括三個步驟︰藉由彩色資訊將格線抽出，去除雜訊，以及偵測格 線。第二部分，彩色地圖先被轉成灰階影像。我們使用局部臨界值法（ local thresholding）來分割出地圖的前景（foreground）。這整個過程 稱為二值化（binarization）。接著，細線化（thinning）被用來獲得前 景的骨架（skeleton）。我們提出一個改良的細線化方法來做線段抽取的 工作。因為地圖的品質不佳及二值化的方法未臻完美，細線化的結果包含 細小及破碎線段。我們必須去除細小線段並連接破碎的線段。在描述道路 之前，我們將地圖的外框去除。第三部分，我們以中心線描述道路。曲線 先以直線段近似，再補起覆疊字所引起的道路線段缺口。接著，我們利用 相配對的道路兩邊的線段產生它們的中心線段。因為道路本身是相連結的 ，我們可以利用此一特性將產生的道路中心線也連接起來。我們測試樣本 地圖七張，其中包含790段道路及446個道路交叉點。所有的格線皆被成功 的偵測出。道路節段被完整抽出的抽出率是95.8%。另外，道路交叉點的 確認率為94%。 This thesis extracts the information about roads and grid lines automatically from street maps to build a database for user retrieval. First, we use the Hough transform to detect grid lines in a map. The operations include extracting grid lines from the map by using color information, deleting noise, and detecting grid lines. Second, we segment the foreground from the map by using the local thresholding method. The color map are first converted to a gray-scale image. The local thresholding process extracts the foreground in binarization. Thinning is used to obtain the skeleton of the extracted foreground. We propose a revised thinning method for line extraction. Because the quality of a map is not good and the binarization method is not perfect, thinning results contain noise and broken lines. We have to delete short line segments and merge broken line segments. Before the road description, we delete the map borders. Third, we describe the roads. Curves are approximated by linear lines. We concatenate the gaps caused by the overlapped characters with road lines. The center lines of roads are produced from the pairs of piecewise line segments. Since the roads that generate center lines are connective, we can merge two or more end points to preserve the topological connective properties of roads in the final results. In the experiments, seven maps are tested. The test maps include 790 road segments and 446 road junctions. All grid lines are correctly extracted. The road extraction rate is 95.8% and the road junction identification rate is 94%.