隨著科技進步，視覺模擬與數碼城市的應用越趨廣泛且日新月益，然而，現今數碼城市與三維建物模型應用與展示的內容大都提供使用者進行大範圍的瀏覽與使用，而較少提供比較細緻的物件模擬或瀏覽。本研究以OGC LOD2的建物模型作為基礎，提出有效的方法利用近景影像提高三維建物模型的細緻化等級。本研究所研發的方法利用影像特徵以及實際建築物結構元件之特性為判斷條件，並且配合語義學分析與影像處理技術，有效判釋建物的窗戶、門、陽台等建築結構元件。再利用空間前方交會進行三維定位，以多張影像量測門窗等元件的尺寸，進行附屬物件之重建，達成三維建物模型細緻化，以快速、有效率且準確的完成具有較高細節層級的建築模型。 The development of computer graphics and geospatial technologies lead to the fast progress in Cyber City implementation and applications. Due to the complexity and diversified applications of digital city models, different levels of detail (LOD) may be necessary in order to fulfill different requirements. Currently, popular digital city systems, such as Google Earth or Bing Earth, mostly utilize LOD1 or LOD2 models, most of the applications still focus on large-scale use and low level of detail building models, these may be inadequate for applications in the engineering domain and other fields. This research develops systematic methods and procedures to increase the level of detail of 3D building models using close-range images and based on low level of detail building models. The idea is to refine CityGML LOD2 models with detail facade objects extracted from close-range images to increase the level of detail of 3D building models. The developed algorithms use image features and characteristics of building structure objects in real world with semantic analysis and digital image processing to recognize the structure objects like windows, doors, balconies, etc. The developed algorithms employ digital image processing to establish the rules to recognize the structure objects and building details from close-range images. Dimensions and positions of the objects are also determined . The identified objects are then added onto the LOD2 models to reconstruct high detailed model.
