On the performance and consistency of a noninteractive multiple-model filter

Tracking maneuvering target using multiple models is an attractive approach that is an alternative to a design that needs logic for both maneuver detection and filter re-initialization. Common current practice in multiple model tracking uses a switching Markov model. Recently the authors developed a multiple model approach to tracking maneuvering targets, but without using a switching Markov model. The models used work independently, while the process noise of each is adjusted online based on a relative likelihood function. The performance and consistency of the newly developed filter are compared to an IMM tracker.