以光條紋投影輔助雙眼內視鏡影像的三維重建

透過內視鏡影像三維重建，醫師可在進行微創手術時獲得手術區域的深度資訊以及重建後的三維影像平面。在本篇論文中主要針對腹腔微創手術的內視鏡影像進行研究及討論。腹腔內視鏡影像和其他種類影像¬─例如室外或是室內場景等─之間最大的差異在於人體內部為平滑且色調單純的表面，在重建時不容易提取出足夠數量或強健性高的特徵。另外，腹腔為一動態環境，人體自然運動包括呼吸造成的體腔反覆移動或是內臟本身的運動如腸道蠕動等都會影響到重建的準確性。而腹腔內部自然分泌的體液以及手術過程中切割產生的血液會讓手術區域的反射性提高，在影像中產生飽和的區域。而手術過程中手術器械除了造成遮蔽，夾或切割臟器皆會造成目標區域的變形，而且電燒止血產生的煙霧則會造成影像清晰程度下降。以上這些問題都會影響重建腹腔內視鏡影像模型的準確性以及完整性。在腹腔鏡影像重建方法部分可分為單純視覺影像重建或是使用額外的器材輔助兩種。單純只使用視覺影像重建又可分為單眼視覺以及雙眼視覺，而使用額外器材輔助有三大類，包括利用雷射、投影等方式增加特徵點、融合影像和其他深度資訊、以及利用手術器械在目標區域造成的陰影。而在本篇論文中我們提出的方法為利用雙眼視覺加上投影色光條紋輔助內視鏡影像的重建，首先校正雙眼相機以獲得相機參數和校正後的影像，然後透過灰階化以及背景相減獲得影像中粗略的條紋區域，再利用閥值以及中值濾波器濾除雜訊以獲得邊緣平滑且較完整的條紋區域。接著擷取每一組雙眼視覺的影像中投影條紋的邊界後，可直接計算其視差，而其餘區域的視差值可通過內插相鄰邊界的視差值獲得。最後，有效視差值的像素其三維座標則可以用三角測量計算。本論文中實驗的部分，我們採用一人體模型的腹部作為影像的目標區域，並以兩個蛇管內視鏡加上微投影機建置實驗所需的感測器。而在實驗中首先取得數組在不同距離和位置所拍攝的腹部影像，接著分別使用論文中所提出的方法重建其三維模型，並針對參數變化對重建結果產生的影響進行比較與討論，最後比較使用本論文所提出方法或是其他重建演算法所獲得的重建結果。

關鍵字

內視鏡影像；三維重建；微創手術；雙眼相機；光條紋投影

並列摘要

Applying 3D surface reconstruction to acquire the depth information and the reconstructed model of the operating region is essential for the surgeons to obtain clearer and expanded view during minimally invasive surgery (MIS). In the thesis, the focus in on the 3D surface reconstruction of laparoscopy images, the one about abdomen surgery. The main challenges about reconstructing laparoscopy images are lacking features, saturation caused by the reflection of body fluid or blood, deformation of the operating surfaces, image blurring caused by the camera motion or naturally movement of human body or organs, smoke coming from burning, and surface occlusion. In the thesis, the focus is on reconstructing the smooth and textureless surface. In order to deal with insufficient features, the proposed method is using the stereo camera with projected light stripes to create additional features on the target surface. With the calibrated and rectified stereo camera, background subtraction and filtering are applied to the image pairs to extract the regions of the light stripes. After obtaining the edges of the light stripes, curve matching and calculating the disparity at the curves is straightforward. As for the disparity in the regions between adjacent curves, interpolation is applied. Finally, since the disparity map is reconstructed and the camera parameters are known, the 3D point cloud can be obtained by applying triangulation. For the experiment in the thesis, one stereo camera is constructed by two endoscopes and fixed with one pico projector. Then several sets of image pairs about the bowel region of the human body model are captured at different distances and positions. The reconstruction results of these image sets are shown and discussed. Besides, the result is compared with those results from other algorithms.

並列關鍵字

endoscopy ； 3D surface reconstruction ； minimally invasive surgery ； stereo camera ； light stripe projection

參考文獻

[1: Wang et al. 2016]

T. Bergen and T. Wittenberg, “Stitching and Surface Reconstruction From Endoscopic Image Sequences: A Review of Applications and Methods,” IEEE Journal of Biomedical and Health Informatics, Vol. 20, No. 1, pp. 304-321, January 2016

[2: Mountney & Yang 2009]

P. Mountney and G. -Z. Yang, “Dynamic View Expansion for Minimally Invasive Surgery using Simultaneous Localization And Mapping,” in Proceedings of Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pp. 1184-1187, 2009

[4: Wu et al. 2007]

國際替代計量

以光條紋投影輔助雙眼內視鏡影像的三維重建

全文下載

主題瀏覽