作者: Carl Olsson , Olof Enqvist
DOI: 10.1007/978-3-642-21227-7_49
关键词: Mathematics 、 Degenerate energy levels 、 Initialization 、 Pairwise comparison 、 Artificial intelligence 、 Computer vision 、 Bundle adjustment 、 Robustness (computer science) 、 Structure from motion 、 Epipolar geometry
摘要: We present a non-incremental approach to structure from motion. Our solution is based on robustly computing global rotations relative geometries and feeding these into the known-rotation framework create an initial for bundle adjustment. To increase robustness we new method constructing reliable point tracks pairwise matches. show that our can be seen as maximizing reliability of track if quality weakest link in used evaluate reliability. estimate final geometry alternate between adjustment robust version formulation. The ability compute both camera translations independent initialization makes algorithm insensitive degenerate epipolar geometries. demonstrate performance system number image collections.