Abstract
The goal of this book was to review some of the central advances in three decades of research on motion estimation and motion analysis from image sequences—from the early variational approach of Horn and Schunck for 2D optic flow estimation to recent variational approaches for 3D scene flow estimation and extensions. In particular, we discussed several algorithmic strategies for minimizing respective functionals, with particular emphasis on an alternation of correspondence optimization and vector field smoothing. This provided profound insights into the process of apparent motion analysis in image sequences. Based on stereo depth estimation and on extending the 2D optical flow to three dimensions, a real-time capable approach to dense scene flow estimation was presented. We discussed metrics to evaluate if a scene flow vector belongs to a moving object and a method to derive an uncertainty associated with scene flow velocity estimates.