Heterogeneous Gaussian and Student's t Fusion for Distributed Target Tracking

In this paper, the heterogeneous fusion of the Kalman and Student's t filters is considered in the context of distributed filter fusion for target tracking. This problem is involved in a multi-sensor tracking scenario where each sensor runs either a Kalman or Student's t filter and they cooperate with each other via fusing the posterior density in a peer-to-peer fashion. This type of heterogeneous fusion has never been investigated before without closed-form solution. What is more, these sensors/filters are inherently correlated with each other to an unknown degree which raises a significant challenge for robust fusion. To address these challenges, both the arithmetic and geometric average fusion approaches are extended based on the appreciated moment matching strategies, in order to maintain the Gaussian or Student's t distribution of the local posterior. The effectiveness and robustness of the proposed methods are verified through simulations which have demonstrated the superiority of arithmetic average fusion method over covariance intersection fusion and augmented measurement fusion.