Image-based modeling of human gaits with higher-order statistics

We present a novel approach to modeling human gaits such as walking and running. We represent the trajectories of a certain number of salient features on the human body as the output of a dynamical system driven by an unknown stochastic input. We present techniques for inferring model parameters and input signal distributions corresponding to different optimality criteria, and evaluate the corresponding models for accuracy and predictive power. In particular, we exploit the higherorder statistical information content in motion capture data to arrive at input signals with independent components. We show that human gaits synthesized from nonGaussian inputs best capture the dynamic complexities of the original gait data.