Unifying Bayesian networks and IMM filtering for improved multiple model estimation

Multiple model filtering has become an important concept for various applications such as maneuvering target tracking or road vehicle positioning. Algorithms like the interacting multiple model (IMM) filter allow an adaption of the filter bandwidth to different motion patterns of the target. In general, the individual probabilities of each model are derived from the estimation itself and the incorporated measurements. In this paper, an approach to exploit additional uncertain knowledge for multiple model filtering is presented. This method is modeling the additional information in a meta model using a Bayesian network. Thus, two important concepts of information fusion are unified to a holistic approach for multiple model filtering. The proposed method is demonstrated on the example of maneuver recognition for road vehicles.