應用Hough轉換於屋頂模型之重建

Abstract

近年來隨著電腦科技的進步，數值式地圖日趨普及，各領域需求日增，如都市景觀設計(Urban Planning)、電子通訊(Telecommunication)、建築工程設計及環境工程，均可以透過數值空間資訊以虛擬實境的方式代替傳統的作業方式。然而，如果完全以人工的方式建構這些三維的資訊，將耗費相當多的時間與人力。以自動化人造建物三維重建取代人工作業，空間資訊將能更廣泛地提供一般的應用。 在重建三維人造建物過程中，以線特徵萃取演算法所萃取的線特徵只代表灰階變化處，未必是真正的建物邊緣線。多餘的資訊不但會造成空間上和計算上的負擔，也會影響重建的成果，所以剔除多餘的資訊是萃取圖徵過程中必要的。本研究利用建物高於地表的性質，剔除地表像元，對後續的處理上，有些許幫助。同時，應用Hough 轉換，本研究提出屋頂重建的一個架構。 為了測試本研究所提方法的可行性，選用ISPRS的Suburb Dataset，其影像內容為位於郊區獨棟之建物，週圍有樹及花園，樓層高度為一到三層，選取其中二塊影像為測試影像，複雜度比整幅影像來的小，可重建出部分成果。但因為資料量不足，所以仍有缺漏的情況發生。

With the progress of the computer technology, digital map is becoming popular and demanded by many disciplines such as Urban Planning, Telecommunication, Architecture Design and Environmental Engineering. They can take use of virtual reality to enhance the traditional operations with digital information. However, these 3D information are acquired by human, the expense is high both in time and human resource. With the automated reconstruction of man-made building, spatial information can be applied widely in many fields. During the process of 3D man-made building reconstruction, linear features extracted with feature-based extraction algorithms only stand for the places of abrupt grayscale change, not necessarily the true edge of building. Building information mixed with others not only increases storage and computation burden, but also degrades the reconstruction. It is necessary to remove redundant information during feature extraction. This research uses the property that the building is higher than the terrain surface to remove the ground pixels. A scheme for reconstruction of man-made building with Hough Transform is proposed in this study. In order to test the proposed scheme, two subscences are extracted from ISPRS’s Suburb Dataset as test images. The experiment shows that only the partial 3D building boundary can be extracted due to insufficient information.