Distributed Multi-sensor Estimation Fusion Based on Minimizing the Jeffrey Divergence Sum

In distributed multi-sensor estimation fusion, the correlations between local measurement noises are typically unknown. Compared to transmitting the estimations and error covariance matrices, transmitting the posterior density functions from local sensors to the fusion center provides more information. Therefore, we model estimation fusion as the minimization of the statistical distance sum between the fused density and local posterior densities. In particular, we formulate the corresponding optimization problem based on the Jeffrey divergence and analyze the convexity of the objective function. The solution of this optimization problem (fused estimation) can be acquired by block coordinate descent algorithm. Numerical experiments demonstrate the effectiveness and conservativeness of our method, which outperforms several other fusion methods in certain scenarios.