| 摘要: | 光達是近年來一個受重視的新光學遙感探測技術。對於三維模型重建而言，雷射掃描點雲資料具相當價值，能細緻地重建建物表面，如牆面、屋頂面、柱子、窗戶等細部的物件特徵。由於點雲資料為一群離散的三維座標點群，難以直接利用原始資料中所隱含的三維資訊，因此在產生三維建物模型前，必須先經過前處理，例如：組織不規則三角網、點雲資料區塊化、結構特徵辨識、結構特徵萃取等等。 以光達點雲重建三維房屋模型可達高細緻度層級，但由於獲取資料形式之限制，欲得到精細準確的三維模型，重建過程繁複。本研究以空載及地面光達點雲為資料來源，重建符合 OGC CityGML LOD3 層級規範的三維房屋模型。過程主要分為三個部分：光達點雲資料套合、點雲群組化及三維表面重建。首先對目標測區獲取空載及地面光達點雲資料後，以七參數轉換方式對空載及地面點雲進行套合。接著根據點雲之幾何條件如共面條件等，將點雲群組化，使同群點雲具相同幾何條件。接著，以不同策略針對平面點群及曲面點群重建三維表面，最後整合重建結果產生完整三維模型，並敷貼高解析紋理影像於模型表面。重建之模型並以地面實測方式驗證模型精度。 LIDAR, Light Detection And Ranging, is an emerging technique for optical remote sensing recently. Laser scanned point cloud is a valuable data source for building reconstruction because it can recover detailed building façade structures like wall, roof, pillar and window etc. Since the point cloud data are discrete, it is difficult to acquire useful 3D information directly. Therefore, before the generation of building models, necessary pre-processing, such as triangulated irregular network (TIN) organization, segmentation, or feature extraction etc., must be carried out. Modeling from point cloud data can achieve high level of detail. However, due to the limitation of data, there will be complex procedures to reconstruct a highly accurate model. This research aims to reconstruct building models conforming to OGC CityGML LOD3 standard from LIDAR point clouds. The proposed method is divided into three main parts: data registration, points partitioning and surface reconstruction. First, after acquiring point cloud data from airborne and ground-based LIDAR, they are merged to a single dataset using 7-parameter transformation. The merged point cloud data are then partitioned into several groups according to different conditions such as coplanarity etc. For each point group, a three-dimensional surface is constructed based on Least Squares Method. Finally all surfaces are merged to reconstruct 3D models, and high resolution images are used as façade texture. The accuracy of reconstructed models are evaluated with ground measurement of check points. |
