數碼城市的應用在近年內越來越普及，且廣受各界的注重。然而，針對數據和圖資的取得管道與高額的費用，以及如何開發有效率的三維建物模型建製之標準作業程序，皆需去克服與改進。本研究所提出的演算法能利用網路上提供的完全免費的數據（包含衛星、航照與街景影像）來建製三維建物模型。此特性將有助於部份資源有限的國家或地區，在現有圖資的條件下建製出基礎的三維建物模型。為了達成此目標，本研究提出了一個融合各類影像的作業流程，主要分為兩個步驟：1.建築物辨識：藉由航遙測影像偵測出建物的外輪廓線；2.不同細緻成度的三維建物模型建造。第一階段包括影像增揚、分類、邊緣檢測、影像形態演算法、向量化與規則化;第二階段則由平面區塊建立出積木模型(BlockModel)，並針對著名地標建立包括紋理敷貼的高細緻度模型。研究結果顯示所提出的方法能成功且有效的利用免費的圖資與數據建製三維建物模型，在測試區1224棟建築物影像當中，成功偵測出1150棟建築物，達到93.94％的完成度。重建的三維模型在平面的位置有相較合理的準確度。本研究利用開放資料及影像，開發出相對低成本的三維建物模型建製程序，其成果適用於都市規劃或防災演練等領域當中。In recent years, cyber cities have become very popular and useful in many applications.However many problems remain unaddressed, specially those related to the difficult andexpensive acquisition of the data required to create three dimensional (3D) building modelsand the tedious and time consuming procedures to be carried out.The objective of this study is to propose a procedure for creating 3D building models usingcompletely free data available on the internet including satellite and ground images. Theproposed approach can be used as a framework for building spatial datasets (3D buildingmodels) for countries or regions with limited access to advanced commercial and other spatialdata sources. In order to achieve this goal, a combination of different image processingalgorithms were performed. The entire process was divided into two main parts: buildingsdetection, in which the purpose is the determination of the building outlines of the citythrough an object-based analysis, and the creation of the 3D building model in different levelsof detail (LOD). The first phase comprises tasks of image enhancements, classification, edgedetection, mathematical image morphology algorithms, vectorization, regularization andrefinement. The second phase includes the building modeling from floor plans, and thecreation of highly detailed models for a few landmark buildings, including textures mapping,selection, croppig, rotation and inpainting.The final results prove that the proposed approach is effective for creating 3D buildingmodels using free data. A quantitative analysis indicates that 1150 buildings out of the 1224buildings shown in the image were successfully extracted, achieving 93.94% of completeness.The reconstructed 3D building models are also accurate in terms of location. The methodproposed in this study provides a relatively economical alternative for creating 3D citymodels from freely available images and other data. The generated models can be used fordifferent applications such as city planning and location-based service.
