Multiple Model Filtering with Switch Time Conditions

Abstract

The interacting multiple model filter has long been the preferred method to handle multiple models in target tracking. The filter finds a suboptimal solution to a problem, which implicitly assumes that immediate model shifts have the highest probability. We argue that this model-shift property does not capture the typical nature of maneuvering targets, namely that changes in target dynamics persist for some time. In this paper, we propose an adjusted switch time assumption that forces the dynamic models to remain fixed for a specified time. The modified filtering problem has lower complexity, and we derive a state estimation algorithm that is close to optimal in many scenarios. From Monte Carlo simulations, the new filter is found to yield a 20% decrease in root mean square position error, compared to the interacting multiple model filter in situations where the switch-time conditions are fulfilled.