以資料蒐評法和影像分塊技術自動萃取空載光達資料中的建物共面屋頂點

Automatic Extraction of Roof Points from Airborne LIDAR Data by Data Snooping and Image Segmentation Technique

比起傳統的二維地理資訊，人們能由三維地理資訊中更有效率獲取更多的空間資訊。而建物模型是三維地理資訊中最基本且重要的一項資料。利用建物三維模型，除了可以在電腦中建立虛擬城市之外，亦可結合不同的資訊系統，進行各種不同的應用。如結合消防單位的防災資訊系統，可將救災工作做到最迅速確實，將災變損失減到最低。而空載光達測量技術是目前能自動獲取大區域範圍高精度且高密度地表資料的新興測量技術，獲取的資料是大量分布於地表的三維點位資料，隱含著空間中有意義的點、線、面特徵，尤其是面特徵，對於自動建構建物模型相當有用，因此空載光達資料變成目前自動建構建物模型的一項重要資料來源，所以由空載光達資料中萃取並建構都市區或含建物區的建物模型供後續應用是目前相當重要的研究主題。而由包含大量精確平面特徵的空載光達資料中，進一步處理萃取有意義的面特徵，甚至是萃取更有意義的地物資訊供後續應用則需要發展不同的演算法。本文即在假設地表建物屋頂面是由三維平面所構成的前提之下，利用建物高程一定高於地面的物空間知識，由包含建物區範圍內的光達資料中，利用影像分塊的處理技術先行萃取可能涵蓋屋頂面的區域資訊，接著將所獲取的建物區域資訊與空載光達資料進行資料融合之後，利用最小二乘平面擬合的資料蒐評法（data snooping）排除不屬於共屋頂面上的光達點資料，並進行自動萃取屬於建物共屋頂面上的光達點供後續精確決定三維建物模型之用。而由實驗結果中證實了本研究所提方法的可行性。 People can acquire spatial information more efficiently and benefit much more from 3-D geographical information than from traditional 2-D geographic information. In 3-D geographical information, building models are essential elements. Not only can building models be used to construct the virtual cities on the web for visualization, but it can also be used as the bases for different information systems. For example, the information system for the precaution or preparedness against natural calamities can be built based on the building models for relieving the victims of a disaster quickly and reducing the damage. Airborne LIDAR surveying systems can acquire automatically high accurate and reliable terrain surface data covered large area. Therefore it is an important source for automatic reconstruction of building models. Basically, airborne LIDAR data are consisted of a large number of 3-D points on the terrain surface; co-planar points should be extracted for subsequent application. Therefore, lots of algorithms are developed to extract the meaningful 3-D co-planar features. Based on data snooping theory, this study proposes an algorithm to extract building roof points from airborne LIDAR data for building reconstruction. Based on the assumption that building roofs are composed of either horizontal or oblique planes and that the heights of roofs should be higher than ground surface, image segmentation algorithms are employed first to segment the possible regions covered roofs from airborne LIDAR data. Based on the theory of data snooping, least square plane fitting algorithm are developed to remove non-roof points and extract the coplanar roof points by fusing those detected roof region outlines for subsequent accurate building reconstruction. From experiments, it also proves the feasibility of the proposed algorithm.