Periodic Motion Extraction Using Harmonic Regression and Structural Parameterization
Abstract: The work of this thesis focuses on how to construct a smooth and periodic motion from motion capture data of a motion with a periodic structure, such as walk cycles. To identify periodic structures in the motion capture data we rely on previously developed methods from algebraic topology to project each frame of the motion capture data onto a circle. This gives us a description of how the periodic motion looks structurally. To construct a typical periodic motion from this projection we propose a harmonic regression model on the positional joint rotations. To prevent overfitting of the regression model we use the joints velocities and accelerations as constraints, which helps in maintaining a natural flow for the motion. Our method helps in preventing overfitting of the regression model, but the extracted motion can still suffer from temporal artefacts due to difficulties in performing the inverse mapping from structure to time. The method described is implemented as part of a plugin for Blender to provide a tool for performing periodic motion extraction in a 3D animation tool-kit.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)