Characteristic errors of the IMM algorithm under three maneuver models for an accelerating target

As a state-of-the-art algorithm, the interacting multiple model (IMM) estimator is widely used for maneuvering target tracking. However, there still is uncertainty in how to design the multiple models used by an IMM estimator. In this context, this paper compares three maneuver models, namely, variable process noise, variable state dimension, and discrete acceleration inputs via computer simulations. The first two modeling techniques belong to the design approach using multiple bandwidths whereas the latter technique utilizes matched filters. The three models exhibit characteristic errors throughout a maneuver, which are termed n-, L-, and u-shaped, respectively. Compared to total root mean squared (RMS) errors, a time-equalized RMS (eRMS) may be more appropriate for a maneuvering target. The total and equalized RMS values are illustrated for the IMM under the three maneuver models for performance evaluation and filter tuning.