Consistent fusion method with uncertainty elimination for distributed multi-sensor systems

Abstract

In this study, the elimination of correlated errors with an unknown correlation in distributed fusion is investigated, and a consistent fusion method for distributed multi-sensor systems is proposed. Unlike most existing fusion methods, the proposed method guarantees the consistency of fusion results without requiring system model parameters or adopting conservative strategies. First, a universal bijection is used to quantify the uncertainty in the estimates to be fused based on the entropy of the independent scalars. Second, the correlated errors caused by unknown mutual information and common process noise are treated as avoidable uncertainties. The avoidable uncertainty is then estimated by using a similarity function based on the Kullback–Leibler divergence. Finally, the avoidable uncertainty is separated from the fusion results by employing a conditional probability model to avoid correlated errors. This method is proven to be unbiased, consistent, and more accurate than the well-known covariance intersection method and the inverse covariance intersection method. The simulation results further verify the superiority of the proposed method in terms of the consistency, accuracy, and ability to limit cumulative errors in sequential fusion processes.