透過您的圖書館登入
IP:52.87.200.112
  • 學位論文

UAV航拍影像點雲產生DSM之研究

The Study on DSM Generated from Point Clouds by UAV Image

指導教授 : 黃金聰
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


無人飛行載具( Unmanned Aerial Vehicle, UAV )因為機動性高、飛行高度低,同時可快速安全的獲取地面上高解析度的影像,現今常用在防救災的規劃與環境監測。但因UAV飛行的穩定度不如傳統固定翼航拍,所以拍攝時影像之間旋轉角度常差異較大,在影像匹配上較不容易。由於尺度不變特徵轉換 (Scale-Invariant Feature Transform, SIFT)具有尺度不變特徵的特性,可以克服拍攝時角度、亮度、像片解析度不一致的問題,故本研究採用SIFT進行特徵點萃取與匹配,以建立相鄰影像的相對關係,再運用移動獲取空間結構 (Structure from Motion, SfM)的方法,解算與建立相機與所拍攝物體之間的空間結構關係。   Photosynth為Microsoft在2008年開放使用的軟體,即運用SIFT與SfM的方式建立每張像片之間的空間關係,並產生具有三維坐標與色彩資訊的點雲資料。本研究以台北大學三峽校區宿舍旁,植被區與土石區這兩塊土丘為實驗區進行UAV航拍,將航拍影像以傳統航空攝影測量方法產生的數值地表模型(Digital Surface Model, DSM)為高程精度比較的依據,與Photosynth點雲產生的DSM進行比較。精度指標為兩組DSM資料高程的RMSE值,透過精度指標的計算,以了解Photosynth點雲產生地表高程的可達精度,並分析植被與土石這兩種不同地表覆蓋時,對高程精度的影響。   由實驗可知,以航空攝影測量自率光束法解算的檢核點精度中,x、y、z方向精度分別為±0.032m、±0.025m、±0.098m,換算為航高比值為1/1377,解算的精度與文獻回顧數據(1:500~1:2600)相仿,故可以做為後續DSM高程的比較依據。另外,以Photosynth產生DSM高程的實驗結果顯示,植被區高程的RMSE值為±0.5050m,裸露礫石的土石區高程的RMSE值為±0.6086m,就本研究所使用的影像而言,當地表覆蓋物不同時以Photosynth方法獲得的高程確實有所不同,其中,以植被區的高程精度優於裸露礫石區的精度。

並列摘要


Unmanned Aerial Vehicle (UAV) has high mobility and low flight altitude char-acteristic so it can take high-resolution images faster and safer, usually used in disasters prevention, protection, and environmental monitoring. However, UAV flight stability degree worse than traditional fixed-wing aerial photography, lead to generate large rotation angle between images, cause images matching not easier. Scale-Invariant Feature Transform(SIFT) is robust and invariant for spatial scale, rotation angle or brightness of the image. Therefore this study uses SIFT for feature point extraction and build the relative relationship of the photos, then use Structure from Motion (SfM) analysis the relative or absolute orientation of several photos, build the spatial structure between the camera and shooting objects.   Microsoft released a free software of Photosynth in2008. It used SIFT and SfM theory to build the coordinates of the object in 3D modeling, with color information. This study choose two of mound near the dormitory which at National Taipei University in Sanxia as experimental area. The aerial photography image used traditional aerial photohrammetry assessment by generated Digital Surface Model(DSM) which as the reference of accuracy Photosynth 3D point cloud genered DSM. The accuracy index of values were adopted for two DSM data. Based on the RMSE of two DSM data, the accuracy of Photosynth point cloud generated DSM would be estimated. As well as analysis the land cover of influence in DSM accuracy.   The result of experiment show, aerial photogrammetry use Self-Calibration methods count, the accuracy of x, y, z is ±0.032m, ±0.025m and ±0.098m respectively. Conversion ratio of flying height is 1/1377, based on the research the accuracy satisfied, can be as the reference of DSM accuracy assessment. In addition, in the case of grass cover mound DSMZ ±0.5050m while the mound cover with gravel shows DSMZ value of ±0.6086m. In this study, land cover on the mound has indeed different at height accuracy, and the mound covered with grass of height accuracy is better than the mound covered with gravel.

並列關鍵字

UAV DSM SIFT Point Cloud

參考文獻


施錦揮、游政恭、鄒慶敏、蔡季欣、林志清、林燕山,2010,無人飛行載具應用於防救災圖資供應之研究--以北二高崩塌地為例。地籍測量,29卷(3期),頁17-36。
吳究、張奇、劉銘哲,2009,基於尺度不變特徵點轉換之精密多元影像套合。航測及遙測學刊,第十四卷第二期,頁141-155。
龔健彬、史天元,2000,Kodak DCS210數位相機量測特性之探討。航測及遙測學刊,5卷(4期),頁39-54.
蕭震洋、謝寶珊、冀樹勇,2011,應用非常規攝影量測評估國道3號3.1公里崩塌事件之土方量。中華水土保持學報, 42(2),頁120-130。
陳思翰,2011,未校正影像三維模型建構與定位精度之研究。國立台北大學不動產與城鄉環境學系碩士論文。

被引用紀錄


黃美甄(2014)。地面控制點對無人飛行載具數值地形模型精度影響之評估〔碩士論文,國立臺北科技大學〕。華藝線上圖書館。https://doi.org/10.6841%2fNTUT.2014.00927
申承翰(2013)。無人飛行載具影像數值地形模型建置及精度評估〔碩士論文,國立臺北科技大學〕。華藝線上圖書館。https://doi.org/10.6841%2fNTUT.2013.00164
蔡欣穎(2015)。無人載具產製向陽國家森林遊樂區地表地形之研究〔碩士論文,國立屏東科技大學〕。華藝線上圖書館。https://doi.org/10.6346%2fNPUST.2015.00249
朱宗威(2014)。無人空中載具數位影像應用於林分蓄積量之推估〔碩士論文,國立屏東科技大學〕。華藝線上圖書館。https://doi.org/10.6346%2fNPUST.2014.00223
翁婕晞(2013)。應用多視角影像於UAV航拍遮蔽區之地形重建〔碩士論文,國立臺北大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0023-3107201317390600

延伸閱讀