Distributed Fusion of Multiple Model Estimators Using Minimum Forward Kullback-Leibler Divergence Sum

The problem of distributed fusion of Gaussian mixture models (GMMs) provided by the local multiple model (MM) estimators is addressed in this paper. Taking GMMs instead of combined Gaussian assumed probability density functions (pdfs) as the output of local MM estimators can retain more detailed (or internal) information of local estimations, but the accompanying challenge is how to perform the fusion of GMMs. For this problem, a distributed fusion framework of GMMs under the minimum forward Kullback-Leibler (KL) divergence sum criterion is proposed first. Then because the KL divergence between GMMs is not analytically tractable, two suboptimal distributed fusion algorithms are further developed within this framework. These two fusion algorithms all have closed forms. Numerical examples verify their effectiveness in both computational efficiency and estimation accuracy.