標題: 不同時期空載光達成果比對探討–以新竹地區為例
Accuracy Validation of the Point Clouds from Airborne Lidar Collected in Two Missions of Different Year in Hsin-Chu Area
作者: 吳紹禎
Shao - Chen Wu
史天元
Tian-Yuan Shih
土木工程學系
關鍵字: 空載光達;高程比對;內部精度;外部精度;平面比對;線形萃取;airborne laser scanner;height accuracy;internal consistency;external accuracy;planimetric accuracy;linear feature extraction
公開日期: 2005
摘要:   本研究主要目的在於比對2002年農委會(COA)與2005年內政部(MOI)二時期的空載光達成果,分析其高程與平面方向偏移量,並探討影響點雲位置的原因。   高程部分,以新竹地區不同時期獲得之空載光達雷射掃瞄資料(2002LiDAR、2005LiDAR)、RTK量測之地面檢核點資料(2004GCPs、2005GCPs)、以及使用2004GCPs進行高程改正後的c2002 LiDAR和經無地面點控制之航帶平差後的c2005LiDAR,進行其相互之間高程偏移量的比對與探討。比對方式分內部精度與外部精度,內部精度為航帶重疊區點雲高程資料的吻合程度,外部精度則為點雲與地面檢核點的高程較差。內部精度成果顯示,不論使用何種改正方式,整體內部精度皆能有所提昇;而外部精度的部分,則透過二不同時期光達資料相互比對驗證,結果顯示2004GCPs含有橢球高轉換不正確所致的系統性誤差,經重新解算修正後,比對成果顯示該系統誤差問題已改善。另外本研究中亦顯示出,使用不同時期空載光達點雲與不同時期之地面檢核點進行比對時,比對成果會受到一些因量測時期地形變化之不同而產生的誤差。   平面部分,使用空載光達原始掃瞄資料(2002LiDAR、2005LiDAR),以高度門檻值分類建物點雲,網格化成50cm平面解析度的數值高程模型,透過Canny邊緣偵測方式萃取建物邊緣資訊,再疊合1/1000線繪圖比對其偏移情形。本研究以自訂的一套流程進行建物邊緣萃取,萃取完成的邊緣線資料分別以視覺疊合方式及誤差距離量測統計方式進行評估。另外對於房屋邊緣角點等邊緣萃取失真處,則使用霍夫轉換法找出光達資訊中較可靠的建物角點進行比對探討。統計成果顯示,本方法評估出空載光達成果整體E向及N向的平面誤差量介於0.259m~0.795m之間,該量級約為三倍高程誤差,與儀器規格相當。另外,由直線偵測方法萃取的建物角點與1/1000線繪圖比對的誤差統計結果顯示,其誤差量較建物邊緣萃取成果評估的誤差值大,這可能是因為建物角點的檢核數較少,因此本研究中建物邊緣萃取的誤差距離量測統計結果應較為準確。本階段研究成果顯示,2002LiDAR建物邊緣的萃取成果有東向的偏移,而2005LiDAR建物邊緣的萃取成果則有北向的偏移。整體上,二時期的評估成果均顯示垂直於航向方向之平面精度優於平行航向方向之平面精度。
 This study evaluates height and planimetric accuracy of the point clouds collected by two airborne lidar missions in 2002 and 2005 respectively. Ground Check Points were collected utilizing real time kinematic GPS in 2004 and 2005 by a field survey team. After analyzing the differences between these four data sets, it is found that there are significant height offsets between two different flight line groups of the 2002 lidar point clouds. Systematic errors in the 2004 GCPs are also identified. With strip adjustment, the internal consistency of the 2005 lidar data can be improved. However, the external accuracy with the GCPs does not show any improvement.  In planimetry, the building boundaries are extracted from the point clouds and compared to a 1:1000 vector map. Hough transform is applied to find corner points for buildings. The errors are then assessed, utilizing both visual and statistical approaches. Experimental results indicate that mean errors were between 0.259m-0.795m in planimetric offsets. The results also show that planimetric offsets in the scanning direction are smaller than those in the flight direction.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT009316577
http://hdl.handle.net/11536/78703
Appears in Collections:Thesis


Files in This Item:

  1. 657701.pdf

If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.