以光達點雲重建屋頂面模型之半自動化作業平台

A Semi-automatic Platform for Building Roof Reconstruction by Using LIDAR Point Clouds

近年來，以光達（LIght Detection And Ranging, LIDAR）資料重建屋頂面模型的相關研究為數眾多，其各自的演算法及自動化程度因考量不同目的或品質成果而有所不同。為了兼顧模型重建的可靠度、效率以及品質，本研究從實務的角度出發進行三維屋頂面重建，工作分成四個階段：首先，從網格化的光達點雲資料中藉由影像處理技術自動量測並輔以人工挑選與修補獲致可靠但資料品質較差之三維結構線（以下稱近似三維結構線）；第二階段則針對近似三維結構線，以先前所發展的建構－－成型演算法重建初始屋頂面模型，並建構組成屋頂面之各個多邊形；在第三階段改良程序則藉由光達離散點雲資料進行平面擬合計算，再將相鄰兩平面進行交會以獲得改良後的三維結構線；第四階段則將改良後的三維結構線（屋頂面內部）與近似三維結構線（屋頂面外緣）進行整體平差計算，以獲取屋角點三維坐標及標準差。實驗結果顯示，本工作所擬定的重建演算策略能獲致幾何精度較高且結構完整的屋頂面模型。 Building roof reconstruction by using LIDAR ((LIght Detection And Ranging)data hasbeen a research focus seen in many studies, varying in strategies, algorithmic designs, levels ofautomation and quality of the product, however. To optimize among reliability, efficiency andquality of the building roof reconstruction, the authors present an effective strategy anddevelop a workable platform of LIDAR data processing for this task. There are four mainprocedures involved in this study. First, rough but reliable initial 3D structure lines areextracted from LIDAR point clouds by combining automatic and manual measurements. Then, the initial building roofs are reconstructed on a 3D structure line basis throughConstruct-Shape algorithmic procedures and followed by constructing polygons of eachbuilding roof. The third procedure is aimed to refine the 3D structure lines that situate insideroofs by going through the following two geometric inferences: (1) Performing the plane fittings; (2) Calculating the intersections of adjacent planes. The final procedure is to adjust theroof geometry by combing the outer initial 3D structure lines and the refined inner 3Dstructure lines. Experiments show that the proposed strategy as well as the developed platformnot only supports a reliable and complete building roof reconstruction task but also exploits thebest geometric quality of LIDAR point clouds.

Building roof reconstruction by using LIDAR ((LIght Detection And Ranging)data hasbeen a research focus seen in many studies, varying in strategies, algorithmic designs, levels ofautomation and quality of the product, however. To optimize among reliability, efficiency andquality of the building roof reconstruction, the authors present an effective strategy anddevelop a workable platform of LIDAR data processing for this task. There are four mainprocedures involved in this study. First, rough but reliable initial 3D structure lines areextracted from LIDAR point clouds by combining automatic and manual measurements. Then,the initial building roofs are reconstructed on a 3D structure line basis throughConstruct-Shape algorithmic procedures and followed by constructing polygons of eachbuilding roof. The third procedure is aimed to refine the 3D structure lines that situate insideroofs by going through the following two geometric inferences: (1) Performing the plane fittings; (2) Calculating the intersections of adjacent planes. The final procedure is to adjust theroof geometry by combing the outer initial 3D structure lines and the refined inner 3Dstructure lines. Experiments show that the proposed strategy as well as the developed platformnot only supports a reliable and complete building roof reconstruction task but also exploits thebest geometric quality of LIDAR point clouds.