AuthorJang Heon Kim
TitleThe Robust 3-D Sceneflow
TutorProf. Dr.-Ing. Thomas Sikora
AbstractRecovering 3-D information from several 2-D images is one of the most important topics in computer vision. Multi-stereo and Structure-from-Motion methods aim to recover the 3-D camera pose and scene structure for a rigid scene from an uncalibrated sequence of 2-D images. The 3-D camera pose can be estimated as the principle projection ray for a camera observing a rigid scene. The estimation of dense scene geometry is a process to recover the metric geometry and to adjust the global ray projection passing through each 2-D image point to a 3-D scene point on real object surfaces. The generalization of 3-D video analysis depends on the density and robustness of the scene geometry estimation. In this dissertation, the 3-D sceneflow method that analyzes jointly stereo and motion is proposed for retrieving the camera geometry and recontructing dense scene geometry accurately.
Key wordsglobal ray projection, 3-D video analysis, sceneflow, camera motion, 3-D sceneflow, stereo-motion sequences