Gaussian Process Fusion for Multisensor Nonlinear Dynamic Systems

This paper investigates the Gaussian process for the problem of the multisensor data fusion with a nonlinear dynamic system. It is well known that the Gaussian process is a nonlinear non-parametric Bayesian method, which is successfully applied to machine learning. Nevertheless, it is rarely used in target tracking for the multisensor data fusion. In order to overcome the difficulty caused by the multisensor nonlinear dynamic systems, we associate the nonlinear transition and measurement function with the Gaussian process regression models, then the advantages of the non-parametric feature of the Gaussian process can be extracted fully for state estimation. Moreover, for the different fusion architectures, we present a centralized fusion method and an distributed fusion method to utilize the information contained in multiple sets of data. Finally, the equivalence between the two proposed fusion methods are established, which is helpful for the research of the multisensor data fusion. A nonlinear target tracking system with two sensors is given to show the effectiveness of the proposed methods.