Engell-Nørregård, MortenHauberg, SørenLapuyade, JeromeErleben, KennyPedersen, Kim SteenstrupHartmut Prautzsch and Alfred Schmitt and Jan Bender and Matthias Teschner2014-02-012014-02-012009978-3-905673-73-9https://doi.org/10.2312/PE/vriphys/vriphys09/077-084We present an application of a fast interactive inverse kinematics method as a dimensionality reduction for monocular human motion estimation. The inverse kinematics solver deals efficiently and robustly with box constraints and does not suffer from shaking artifacts. The presented motion estimation system uses a single camera to estimate the motion of a human. The results show that inverse kinematics can significantly speed up the estimation process, while retaining a quality comparable to a full pose motion estimation system. Our novelty lies primarily in use of inverse kinematics to significantly speed up the particle filtering. It should be stressed that the observation part of the system has not been our focus, and as such is described only from a sense of completeness. With our approach it is possible to construct a robust and computationally efficient system for human motion estimation.Categories and Subject Descriptors (according to ACM CCS): Computer Graphics [I.3.7]: Animation - Computer Graphics [I.3.5]: Physically based modeling - Numerical Analysis [G.1.6]: Constrained Optimization - Numerical Analysis [G.1.6]: Nonlinear Programming - Image Processing and Computer Vision [I.4.6]: Scene Analysis - Probability and Statistics [G.3]: Markov processes - Keywords: Inverse Kinematics, Nonlinear, Joint Limits, Constraints, Motion Estimation