Partial State Interacting Multiple-Model Algorithm for Tracking of Targets With Maneuvering Parameter Jumps

The primary challenge in maneuvering target tracking arises from the uncertainty in mode types and the uncertainty in maneuvering parameters, the latter of which is particularly intractable since maneuvering parameters are typically unknown and may undergo abrupt jumps within a continuous space regardless of whether the mode type is switched or not. In this article, a partial state interacting multiple-model (PSIMM) algorithm is presented for tracking of targets with maneuvering parameter jumps. In the PSIMM algorithm, maneuvering parameter jumps are described by switchings between models with the same structure. Each maneuvering model is initialized with a Gaussian probability density function, which covers a designated partition in maneuvering parameter space to capture maneuvering parameter jumps. To deal with possible jumps in maneuvering parameters, a partial state interaction strategy is proposed, in which the maneuvering parameter input to each maneuvering model is a weighted combination of the previous estimate from the model itself and the specified parameter initialization, rather than the result of interaction with other models. The proposed partial state interaction strategy not only provides a consistent initialization for maneuvering parameters at instants of maneuvering parameter jumps but also prevents the best-initialized filter from being affected by false maneuvering parameter estimates during maneuver sojourn segments. The performance of the proposed PSIMM is evaluated via four maneuvering target tracking scenarios, and simulation results demonstrate the effectiveness of the PSIMM compared to the state-of-the-art multiple-model methods.