Target tracking filter with unknown time-varying system noise variance

In maneuvering target tracking, to ensure the numerical stability and optimality of Kalman filter, the unknown time-varying system noise variance needs to be estimated adaptively, but the existing methods are mostly limited to systems with stationary or slowly-varying noises. Based on the Sage-Husa adaptive estimator of system noise variance, by introducing the innovation-based detection criterion for filter divergence and using the idea of strong tracking filter, a time-varying scaling factor matrix multiplied by the system noise variance estimate is proposed in this paper, which can restrain the likely divergence of the tracking filter due to the sudden jumps of the system noise variance. Monte-Carlo simulation results show that the new proposed algorithm not only has better numerical stability, but also can improve the tracking precision greatly.