A nonlinear and non-Gaussian distributed fusion based on Rao-Blackwellized particle filtering

In nonlinear and non-Gaussian multi-sensor fusion scenarios, the Covariance Intersection (CI) algorithm is utilized to fuse estimations from distributed sensors, in which targets are commonly tracked by a family of Particle Filtering (PF) algorithm. Furthermore, standard PF can be replaced by Rao-Blackwellized PF (RBPF) based on linear/nonlinear State Space models to produce more accurate means and variances for CI fusion. Unfortunately, the RBPF algorithm fails in conventional radar systems because their observations contain no information about the linear part of target state. To overcome such an issue, a Kalman Estimation based BRPF (KE-BRPF) algorithm is proposed to form a novel distributed CI fusion. In KE-RBPF, the correlation between linear and nonlinear parts of target state is investigated. Benefitting from this investigation, the linear part of target state is correctly tracked based on the nonlinear one. Finally, the simulations verify that our KE-RBPF-CI fusion outperforms other PF-based CI fusions, in terms of means and deviations.