Bayesian Filtering with Online Gaussian Process Latent Variable Models

Abstract

In this paper we present a novel non-parametric approach to Bayesian filtering, where the predic-tion and observation models are learned in an online fashion. Our approach is able to han-dle multimodal distributions over both models by employing a mixture model representation with Gaussian Processes (GP) based components. To cope with the increasing complexity of the esti-mation process, we explore two computationally efficient GP variants, sparse online GP and local GP, which help to manage computation require-ments for each mixture component. Our exper-iments demonstrate that our approach can track human motion much more accurately than exist-ing approaches that learn the prediction and ob-servation models offline and do not update these models with the incoming data stream. 1