以非量測性相機進行近景攝影測量探討

Abstract

三維模型之需求逐日增加，利用非量測性數位相機獲取之目標物影像，得到三維空間資訊，已成為一項極具經濟效益之課題。非量測性數位相機之量測性應用，內方位參數為關鍵因素，本研究選用三套近景攝影測量商用軟體PhotoModeler、iWitness及Lensphoto探討相機率定結果及像距與物距之關係。使用Lensphoto 軟體以近景攝影測量技術生成目標物之點雲，並與地面雷射掃描儀獲得之點雲，探討近景攝影測量之限制。以平行及旋轉多基線攝影之方式，拍攝目標物影像，生成點雲資料並分析其精度。 研究成果顯示，兩次實驗在輻射距離最大之輻射透鏡畸變差之差值：PhotoModeler為11.135μm，iWitness為3.489μm，Lensphoto為1.664μm，PhotoModeler之差值為最大；最大輻射透鏡畸變差之平均：PhotoModeler為336.097μm，iWitness為343.531μm，Lensphoto為320.847μm。相機率定時受率定標涵蓋影像畫面之限制，使鏡頭對焦距離非無窮遠，則率定所得的焦距為像距。其中，由於PhotoModeler相機率定標之尺寸不同，造成攝影物距不同，導致率定所得攝影像距差值達1.04mm。 鏡面雕塑之三維建模難以在近景攝影測量中實現，因鏡面會隨不同拍攝角度而呈現不同之反射影像，造成立體像對無法匹配。載入Lensphoto軟體之影像受目標物涵蓋影像範圍之限制，使匹配難以生成整體模型點雲。以平行與旋轉多基線攝影方式取得目標物影像並經空三解算後生成點雲，平行多基線攝影之控制點及檢核點整體精度以RMSE(Root Mean Square Error)表示，分別為0.0066m及0.0678m，旋轉多基線交會攝影則分別為0.0016m及0.0425m，顯示旋轉多基線交會攝影之精度高於平行多基線攝影方式。在近景攝影測量與地面雷射掃描之精度比較，地面雷射掃描之控制點及檢核點整理精度之RMSE 分別為0.0376m及0.0382m，結果顯示近景攝影測量之精度略高於地面雷射掃描之精度。

The demand of 3D models increased day by day, using non-metric digital camera to capture the image with 3D information, has a great economic benefit. In the non-metric digital camera applications, the interior orientation parameters are very important. This study used three sets of close-range photogrammetry software : PhotoModeler, iWitness and Lensphoto, and analyze the result of camera calibration and the relationship between image distance and object distance. Using Lensphoto software to generate point clouds of objects, analyze the limitations of close-range photogrammetry with the point cloud from ground laser scanner. Finally, analyze the accuracy of parallel multi-baseline and panning and multi-baseline digital close-range photogrammetry. Results show that difference of the largest radial lens distortion on the largest radial : PhotoModeler is 11.135μm, and iWitness is 3.489μm, and Lensphoto is 1.664μm. Average of the largest radial lens distortion on the largest radial : PhotoModeler is 336.097μm, and iWitness is 343.531μm, and Lensphoto is 320.847μm. When object distance is equal to infinity, the image distance is equal to the focal length. Due to the different sizes of targets in PhotoModeler, the object distances are different and the difference of image distance up to 1.04mm. It is difficult that modeling 3D mirror sculpture in the close-range photogrammetry, because mirror will reflect rays in different angles and different reflectance sights, resulting stereopairs not match. Because the softwares of close-range photogrammetry have some restrictions of images, the object is difficult to reconstruct overall 3D model. The two methods of taking images are parallel multi-baseline and panning and multi-baseline, and generate point clouds. The RMSE (Root Mean Square Error) of control points and check points for parallel multi-baseline are 0.0066m and 0.0678m, and for panning and multi-baseline are 0.0016m and 0.0425m. From the result of Terrestrial Laser Scanning, it shows that the RMSE of control points and check points are millimeter. From the result of Terrestrial Laser Scanning, it shows that the RMSE of control points and check points are 0.0376m and 0.0382m. From the result of this research, the accuracy of close-range photogrammetry is slight higher than the accuracy of Terrestrial Laser Scanning.