Variable Structure Multiple Model Estimation Base On Particle Filter

Maneuvering target tracking is one of important problems which receive close attention in many fields. But Nonlinearity, non-Gaussianity and maneuvering often bring difficulty to target tracking problem.Particle filter (PF) is an optimal nonlinearity filtering method which rises in recent years. It frees from the limitation which the stochastic quantity should be Gaussian distribution in solving the nonlinearity filtering problem, and the multiple-model (MM) algorithm is suitable for the case of tracking target with high maneuvering. According to the feature of the PF and MM algorithm, this dissertation carries out research systematically in following two aspects:(1) The research on design of the model-set and the approach of the adaptation.MM estimation includes the design of the model-set and the adaptation algorithm. To solve the problem that the traditional model is only fit for the situation which the model used should accord with the true model, the acceleration model-set is proposed for tracking generic motion. And the traditional adaptive gird algorithm is only applied to 2D coordinate turn system, a modified adaptive gird algorithm is proposed here to tracking generic 2D model system.(2) The research on combination of the PF and variable structure (VS) MM.Some researches on combine PF with MM estimation are appear in recent years. For the good performance of the VSMM, it is proposed here to combine PF with VSMM to solve the tracking problem with Nonlinearity, non-Gaussianity and maneuvering. Some methods of the combination of particles are analyzed, the means and covariance of the model is used to produce particle directly. This approach could reduce computational complexity. The combination of the modified AGMM and PF is also used to tracking target which the model scope is unknown in advance.Simulation results indicate the algorithm which combines the PF with VSMM performances well in computational load and tracking error.