Characterizing Differences in Walking Motions with Conditional Gaussian Process Dynamical Model

Abstract

As motor function improves or declines, the way to control the body changes even for the same class of actions. To investigate what parts of the body movements characterize such changes, it is practical to summarize a high-dimensional time series of body movements into a low-dimensional time series by using the information on the coordinated behavior across different body parts. Significant differences in the summarized motion series should likely originate from characteristic differences in body control. In this study, we apply a conditional Gaussian Process Dynamic model to walking motions and propose a method to analyze the characteristics of body motions based on the dynamics of latent variables derived from a high-dimensional body movement time series.