Skip to Main Content
The goal of this article is to present an effective and robust tracking algorithm for nonlinear feet motion by deploying particle filter integrated with Gaussian process latent variable model and embedded with Markov-switching approach. Training trajectory data is projected from the observation space to the latent space of lower dimensionality in a nonlinear probabilistic manner. In the latent space, particle filter is used to track indeterministic motions of feet. The number of particles are reduced by incorporating learning knowledge as well as temporal information explored by Markov switching model. The simulation results indicate that the proposed approach is able to effectively track feet with relatively different motion patterns, and even under temporal occlusions.