A Novel Multiple-Model Adaptive Kalman Filter for an Unknown Measurement Loss Probability
Wonkeun Youn, Nak Yong Ko, S. Andrew Gadsden, Hyun Myung
Abstract
This article proposes a novel adaptive Kalman filter (AKF) to estimate the unknown probability of measurement loss using the interacting multiple-model (IMM) filtering framework, yielding the IMM-AKF algorithm. In the proposed IMM-AKF algorithm, the state, Bernoulli random variable, and measurement loss probability are jointly inferred based on the variational Bayesian (VB) approach. In particular, a new likelihood definition is derived for the mode probability update process of the IMM-AKF algorithm. Experiments demonstrate the superiority of the proposed IMM-AKF algorithm over existing filtering algorithms by adaptively estimating the unknown time-varying measurement loss probability.