On-demand track-to-track fusion using local IMM inside information

Abstract

The fusion of state estimates from Interacting Multiple Model (IMM) estimators using inside information (mixture estimates and probabilities) is described in this paper. Fusion is performed on-demand, i.e., without conditioning on past track information. The local trackers run IMM estimators to track a target and transmit mode-conditioned estimates and mode probabilities to a Fusion Center. The fused state posterior probability density is a Gaussian mixture whose parameters can be computed recursively. The likelihood functions of the state and mode are derived, yielding consistent data fusion. Simulations show that this method outperforms the fusion of the local IMM estimator outputs both in terms of error during target maneuvers and in the consistency of the mean-squared error.