Mixture truncated unscented Kalman filtering

This paper proposes a computationally efficient nonlinear filter that approximates the posterior probability density function (PDF) as a Gaussian mixture. The novelty of this filter lies in the update step. If the likelihood has a bounded support made up of different regions, we can use a modified prior PDF, which is a mixture, that meets Bayes' rule exactly. The central idea of this paper is that a Kalman filter applied to each component of the modified prior mixture can improve the approximation to the posterior provided by the Kalman filter. In practice, bounded support is not necessary.