Partially Interacting Multiple-Model Algorithm With Maneuvering Parameters Initialized by Gaussian Mixture

Abstract

The challenge of maneuvering target tracking lies in the difficulty of handling the uncertainty of target motion types and the uncertainty of maneuvering parameters simultaneously. In this article, a partially interacting multiple-model (MM) algorithm with maneuvering parameters initialized by Gaussian mixture (GM-PIMM) is presented. In the proposed method, a maneuver mode is collectively described by multiple models with the same structure and allowing for slight changes in maneuvering parameters. Each model is initialized by a Gaussian probability density function with a quantized mean. The models with the same structure operate independently of each other after parameter initialization, and the interaction between models of different structures is preserved to cope with possible mode jumps. The strategy of initializing maneuvering parameters using the Gaussian mixture eliminates both the mismatch of maneuvering models and the deviation of maneuvering parameters. The mechanism of partial interaction rather than all models interacting with each other protects the best-initialized filter from interference from other incorrectly initialized models, resulting in satisfactory estimation accuracy during maneuver sojourn segments. Simulation results demonstrate the superiority of the proposed GM-PIMM in terms of estimation accuracy and computational complexity compared to the state-of-the-art MM methods.