Online Full Body Motion Reconstruction

Abstract

This thesis described the process of creating a full Performance Animation pipeline from motion capture to virtual character animation. In this pipeline, techniques are discussed to detect and handle mislabeled motion capture markers, reduce data noise by means of a Kalman filter, extrapolate and interpolate, as well as skeleton reconstruction and inverse kinematics by means of the Cyclic Coordinate Descent Algorithm. The animations are created using different setups of markers and joint freedom, in order to test the effects markers and joints have on the perceived naturalness and determine the optimal configuration. An alternative way of indicating naturalness is described using objective error metrics, which show a moderately strong correlation to perceived naturalness data obtained through a user case study.