The seminar talk presents geometric foundations and
algorithmic approaches to deal with the 2D-3D pose estimation
problem. The presentation starts with point based pose
estimation, continues with kinematic chains, free-form contours and
surfaces and ends with the recovery of human motion.
Additional topics presented and discussed concern outlier and occlusion
detection, morphing approaches for more
realistic human motion modelling, the coupling of image segmentation
methods with 3D shape priors and a performance evaluation between the
developed marker-less motion estimation system and the marker based
Motion Analysis system