A human motion estimation method based on GP-UKF
Ziyou Wang, Jun Kinugawa, Hongbo Wang, Kazuhiro Kosuge · 2014
A novel human motion estimation method is presented in this paper. The motion of the human is estimated by an Unscented Kalman filter (UKF), in which a nonlinear dynamic model is used to predict trajectory of human. This dynamic model is obtained from sample data by using Gaussian Process (GP) regression. The sample data includes information of body segment posture and trajectory data collected by motion capture system. The GP-UKF can extract the underlying dynamics from the sample data, with which the future non-linear transition can be predicted. The experiment results show that the proposed method has improved accuracy over conventional method.