Researches On High-accuracy State Estimation Methods And Their Applications To Target Tracking And Cooperative Localization

State estimation theory is an important research branch of automatic control field,statistical signal processing field,and information fusion field.State estimation methods represented by Kalman filtering and smoothing have been widely applied to the navigation and positioning,target tracking,signal processing,communications,control,robotics,and space exploration.In the case of linear state-space:model with Gaussian noises,Kalman filtering/smoothing has the highest state estimation accuracy.However,in the applications of target tracking and cooperative localization,nonlinear system model,inaccurate noise statistics,heavy-tailed non-Gaussian noise,and heavy-tailed and skew non-Gaussian noise will degrade the accuracy of traditional Kalman filtering/smoothing algorithms,even lead to divergence.The above problems are also important problems in the basic theory of state estimation.In order to improve the accuracy of state estimation methods in the application environments of target tracking and cooperative localization,this dissertation focuses on the above four problems and proposes a series of improved high-accuracy state estimation methods,and the effectiveness and superiority of the proposed methods are demonstrated in the applications of target tracking and cooperative localization.The main work of this dissertation is as follows:1.Researches on high-accuracy nonlinear Gaussian approximate state estimation methods.Firstly,aiming at the problem of Gaussian approximation for the nonlinear system,this dissertation proposes four new Gaussian approximate filters,including high-degree unscented Kalman filter,high-degree interpolatory cubature Kalman filter,high-degree adaptive embedded cubature Kalman filter,and quadrature point propagation based Gaussian approximate filter,and their relationships and application scenes are also revealed.Then,aiming at the state estimation problem for nonlinear systems with large prior uncertainty and high measurement accuracy,the statistical linearization based recursive measurement update method and Gaussian approximation based recursive measurement update method are proposed,which can improve accuracy through absorbing the measurement information more fully.Finally,aiming at the nonlinear state estimation problem with correlated noises,a new Gaussian approximate filter with correlated noise at the same epoch is proposed by jointly estimating state vector and process noise,and Gaussian approximate filter and smoother with correlated noises at one epoch apart are proposed based on the de-correlating method.Simulations of target tracking demonstrate that the proposed nonlinear Gaussian approximate state estimation methods have better estimation accuracy than existing methods.2.Researches on adaptive Kalman filtering methods.In the adaptive Kalman filtering,the estimate of process noise statistic is a very difficult problem.Aiming at this problem,this dissertation proposes an original idea of estimating the one-step prediction error covariance matrix in the light of the feature that the state estimate is more related to the one-step prediction error covariance matrix,which avoids to estimate process noise covariance matrix directly.Based on this idea,a novel adaptive Kalman filter is firstly proposed using variational Bayesian(VB)approach,in which the state vector and inaccurate one-step prediction error covariance matrix and measurement noise covariance matrix are jointly estimated based on VB approach.Simulations of target tracking illustrate that the proposed VB based adaptive Kalman filter has better filtering accuracy than existing adaptive Kalman filters.In practical applications,for the case that the prior information of noise covariance matrices is unavailable,a new adaptive extended Kalman filter(AEKF)is further proposed based on expectation maximization algorithm,and it is used to address the problem of time-varying and inaccurate process and measurement noise covariance matrices in the cooperative localization of AUV.The experiments of cooperative localization illustrate that the proposed AEKF based cooperative localization algorithm has better localization accuracy than existing cooperative localization algorithms.3.Researches on state estimation methods with heavy-tailed non-Gaussian noises.Aiming at the heavy-tailed non-Gaussian noise characteristic induced by outlier interference in the applications of target tracking and cooperative localization,the posterior Student's t approximation based state estimate method and posterior Gaussian approximation based state estimate method are,respectively,proposed in terms of the level of heavy tail by using Student's t distribution to model process and measurement noises.A new robust posterior Student's t approximation based filtering and smoothing framework is proposed for the case of significantly non-Gaussian heavy-tailed process and measurement noises,and the simulations of manoeuvring target tracking show the proposed methods have better estimation accuracy than existing nonlinear robust state estimation methods.For the case of moderately heavy-tailed process and measurement noises,this dissertation provides analysis and achieves the conclusion that Gaussian approximation to posterior probability density function(PDF)is more reasonable than Student's t approximation,and novel robust Gaussian approximate filter and smoother are proposed by modelling noises as Student's t distributed and approximating posterior PDFs as Gaussian.The experiments of underwater cooperative localization and the simulations of target tracking,respectively,demonstrate that the proposed robust Gaussian approximate filter and smoother have better estimation accuracy than existing methods.4.Researches on state estimation methods with heavy-tailed and skew non-Gaussian noises.In the acoustic range based underwater cooperative localization,the large positive measurement error may be induced by the multipath propagation of sound wave between sound source and receiver,which results in non-Gaussian heavy-tailed and skew measurement noise.In order to solve the state estimation problem with heavy-tailed and skew non-Gaussian noises,a new robust Gaussian scale mixture(GScM)based Kalman filter is proposed by using the GScM distributions to model the one-step prediction PDF of state and measurement likelihood PDF,in which the state vector,mixing parameters,and the scale matrices and shape parameters of GScM distributions are jointly estimated based on VB approach.To improve the modeling accuracy of non-Gaussian heavy-tailed and skew noise,a new generalized Gaussian scale mixture(GGScM)distribution is proposed,based on which a new robust GGScM based Rauch-Tung-Striebel(RTS)smoother is proposed.The experiments of cooperative localization show that the proposed robust RTS smoothing algorithm has better estimation accuracy than existing state-of-the-art RTS smoothers.