作者: Théo Papadopoulo , Olivier Faugeras
DOI: 10.1007/3-540-61123-1_182
关键词: Mathematics 、 Real image 、 Computer vision 、 Motion field 、 Constraint (information theory) 、 Artificial intelligence 、 Visibility (geometry) 、 Motion analysis 、 Cardinal point 、 Perspective (graphical) 、 Optical flow
摘要: This paper discusses the problem of recovering 3D motion and structure for rigid curves. For general curves, there is exactly one constraint each image point that relates so-called spatio-temporal derivatives to instantaneous parameters. However, this directly derives from perspective equations thus ignores fact viewed object must be entirely in front focal plane. provides a description some constraints insuring visibility curve examines their geometrical consequences reconstructions. Ignoring them leads erroneous solutions with high probability even when using perfect data. An algorithm which takes advantage these proposed. The accuracy obtained results demonstrated on both synthetic real sequences.