偵測房屋區塊為重建三維房屋模型重要工作之一，也是遙測重要的研究項目。利用多重影像匹配後之三維點雲建立的幾何條件與影像的光譜資訊，相互整合利於偵測房屋區塊。為使匹配後產生之三維點雲能完整的展現房屋幾何，除了常使用的點特徵外，擬加入線特徵與重複紋理區塊特徵進行匹配。本計畫以三年為期進行高重疊航照影像之多特徵匹配於房屋區偵測。其中，各年度工作目標包含(1) 點特徵與線特徵匹配重建數值地表模型，(2) 整合點、線、重複紋理區塊特徵匹配重建數值地表模型，(3) 多特徵匹配配合光譜資訊之房屋區偵測。第一與第二年工作重點為高重疊影像多特徵匹配，配合影像金字塔與多視窗匹配的概念，提升影像匹配成果可靠度。其中，第一年使用點特徵與線特徵，使用多影像匹配尋找共軛目標，建立數值地表模型；第二年增加重複紋理區塊之特徵匹配重建地表模型。三種不同類型之特徵在匹配過程中，均需滿足核線幾何的約制條件外，線特徵匹配滿足直線對應直線，重複紋理區塊特徵則考量紋理中基本結構元彼此對應的空間關係。第三年研究重點為建立多資訊之房屋區偵測模式，藉由前述處理技術與重建成果分析地表幾何條件，再整合影像的光譜資訊，經由一多面向的條件分類程序，進行房屋區塊偵測。 Building detection is an important task for both 3D building reconstruction and many applications of remote sensing. The 3D point clouds which are obtained from multi-images matching could provide geometrical conditions of objects on the surface. It is helpful to building detection when we combine the geometrical conditions and the spectrum information of images. In order to complete the building shape for 3D points cloud, lines feature and the reduplicate texture areas are employed in the image matching. This three-year project investigates multi-feature matching for high overlap images for the detection of building areas. The milestone for each year is (1) multi-feature matching for points and lines for Digital Surface Model (DSM) generation. (2) Integrated points, lines, and reduplicate texture to multi-feature matching for DSM generation. (3) Detection of building areas with geometrical conditions of multi-feature matching and the spectrum information of images. The major work is multi-feature matching for high overlap images in the first and second year. Image matching with image pyramid and multiple windows is employed in this study to increase matching reliability. In the first year, this project will use point and line features to produce DSM using multi-image matching. It considers the reduplicate texture to integrate the image matching in the second year. In addition to the restriction of epipolar lines, it’s considered that the corresponding line characteristics for line features. The spatial relationship of the basic elements is analyzed in the reduplicate texture area for feature matching. The major work in the last year is to set up a model for building area detection. The analyses for geometrical conditions from DSM and examination for the information of images’ spectrum are performed for building detection. A procedure with multi-information classification is employed to detect building areas. 研究期間：10008 ~ 10107
