Gaussian mixture reduction based on KI divergence
Zhiying Yao, Dong Liu · 2010
A common problem in Gaussian mixture filtering is to approximate a Gaussian mixture by one containing fewer components. Similar problems can arise in integrated navigation and multi-target tracking. The paper proposes a new algorithm based on pairwise merging of gaussian mixture components, but in which the choice of components for merging is based on KI divergence of the post-merge density with respect to the pre-merge density. The behavior of the several algorithms in the literatures is compared using an indicative example. And a measure is defined to evaluate the difference between the reduced mixture and the original mixture. The simulation results indicate that the proposed algorithm outperforms the other four algorithms.