整合三維房屋模型與二維影像之套合、匹配、及牆面重建之研究(III)

Abstract

三維房屋模型是數碼城市及適地性資訊服務中重要的數位內容。本計畫使用都市等 級三維房屋模型(OGC LOD2)及二維近景影像，建立建築等級三維細緻房屋模型(OGC LOD3)。研究重點為三維房屋模型與二維影像之空間套合、特徵匹配及牆面幾何重建。 本計畫年度工作目標為半自動化三維細緻房屋模型之牆面幾何重建，研究重點為牆面幾 何重建及紋理敷貼。研究中使用物件導向知識庫分類偵測牆面之窗、門等物件，以半自 動化方式建立階層式牆面語意模型。牆面語意模型包含牆面物件間之關聯性及屬性。接 著由牆面語意模型建立牆面幾何模型，利用三維線段經模型庫之模型擬合，產生牆面幾 何模型。為避免牆面物件幾何模型間的差異，研究中對所有物件進行整體正規化處理， 以提升牆面物件間的一致性。最後，將影像紋理敷貼於牆面物件，完成建築等級三維細 緻房屋模型重建。

Three-dimensional building model is an important digital content for Cyber City and Location-based Service. This research is to establish an OGC-LOD3 building model from OGC-LOD2 building models and close range images. The core of this research contains space registration, feature matching, and façade reconstruction. The objective of this project is to reconstruct the façade model for OGC-LOD3 building model as well as texture mapping. First, object-oriented knowledge-based classification is selected to classify the window and door objects. Then, we generate the hierarchical grammar model which includes the relations and attributes for façade semi-automatically. A model-driven reconstruction is then applied to fit the 3-D primitives and 3-D structural lines for 3-D objects. In order to refine the consistency between façade objects, geometric regularization is preformed to adjust the size and location of façade objects. The texture mapping is the last step to improve the quality of façade.