A Dense Matching Method with Feature Based Descriptors for Non-Parallel Optical Axes Images

論文翻譯標題: 用於非平行光軸影像之基於特徵描述稠密式匹配法
  • 湯 力行

學生論文: Master's Thesis


3D scanning has been widely applied in different fields because of its rapidity coverage and low cost 3D scanning can be used to scan objects of different sizes and in different fields including manufacturing architecture art/history and healthcare However when using 3D scanning for small sensitive items it is necessary to operate more carefully especially in the case of small human organs Generally a non-contact scanning method is the best choice for such delicate areas of human body The non-parallel optical axes system stereo vision system which is a non-contact scanning method is a feasible method for small items The theory of depth measurement depends on the geometrical relationship between the cameras and the disparities obtained from image matching Instead of placing cameras in parallel as is customary in traditional stereo vision it acquires more information from an image by rotating the optical axes to close to the object However the rotation of optical axes results in changes in the image and the traditional local dense matching method doesn’t perform well when this occurs Thus this thesis provides a modified stereo matching method for the non-parallel optical axes system To overcome the image transformation caused by the rotation of the optical axes a feature based descriptor is taken as the matching cost and combined with local dense matching for the purpose of reconstructing the dense point cloud Then the accuracy is improved by constraining the search area using the modified epipolar constraint of the non-parallel optical axes system Also parallelizing the algorithm makes the proposed method more competitive This research provides a modified image matching method for non-parallel optical axes to scan small human organs
獎項日期2018 十月 2
監督員Ta-Chung Wang (Supervisor)