| 摘要: | 近年來，三維房屋模型已普遍運用於都市規劃、環境與景觀模擬、3D導航、災害防治模擬，以及建築文化遺址的保存…等。尤其對文化遺產而言，有效的應用三維建物資訊，快速並確切展示結構物特色，以提升文化研究與管理維護，為目前的重要課題。建置模型時，有幾種常用的表示法，如：線框骨架、參數、網格模型…等，使用參數式模型能有效地描述最簡易的結構類型；反之，網格模型能詳細地呈現出複雜建物的細節。因此，本研究提出了一個兩階段由粗略到細緻的策略，整合兩種模型的優點，藉由高解析影像點雲資料，建置三維複雜建築結構或考古遺址，以降低資料儲存空間並保有結構物的細節。
整體流程利用數位單眼相機(DSLR)取得目標物之原始影像資料，以運動回復結構(SFM)演算法生成三維點雲。經雜訊濾除後，點雲藉由群組化分類萃取基本幾何參數，配合複雜度計算，將複雜度較低之點雲建置成參數式模型，而複雜度高的點雲以波以松比表面重建成網格模型。整合參數模型與網格模型為本研究的核心目標，共分為三大步驟。首先，偵測網格模型接合面之邊界點，利用其方向向量作為網格模型的邊界條件。將邊界點逐一投影至預接合之參數表面上生成虛擬點，連接邊界點與虛擬點，於預整合空間中形成一轉換面，最後建置成整合模型。本研究測試一組歷史性建物，測試區鄰近中央大學校區。於成果階段將展示整合模型並對於其效能進行分析討論，再透過所開發之整合模型細緻度層級(LOD)，呈現不同精緻度之複雜建築物。
;Nowadays, three dimensional (3D) building model has been widely used in urban planning, landscape simulation, 3D navigation, disaster prevention simulation and preservation of architectural heritage sites. Especially for cultural heritage, applying 3D building information effectively and representing the characteristics of structures fast and accurately are important tasks to enhance cultural research and promote the maintenance of management. There are several representations commonly used for building models such as wire frame, parametric, mesh models, and so on. An effective approach is to use parameterized models describing the most common building types. In addition, another common approach to represent buildings is using meshes. Meshes or triangulated local adaptive meshes can be used to describe complex objects in detail. Accordingly, this research proposes a two-step, coarse-to-fine strategy to create three-dimensional models of complicated architectural structures or archeological sites based on point clouds data. The proposed approach creates a sparse model first and then the fine geometric details will be added gradually.
In this study, digital single-lens reflex (DSLR) camera is used to collect the original image data of targets. Then, 3D point clouds are generated from these images using Structure from Motion (SFM) algorithms. Parameters for constructing parametric models are identified from the point clouds. Moreover, Poisson surface reconstruction is carried out to generate mesh models. The integration of parametric and mesh models is the major subsequent task in this research, which includes three steps. Firstly, detect boundary points for connecting parameter models and mesh models. Next, generate a new junction on the pre-integrated surfaces of the parameter models. Finally, a new mesh model is generated in the pre-integrated space. The result demonstrates a hybrid model of complicated buildings with different Levels of Details (LOD). |
