Research On Variational Bayesian Target Tracking Algorithm Under Uncertain Measurement

With the vigorous development of modern aviation,navigation and aerospace,target tracking has attracted more and more attention from many countries.At present,it has become a very active research field.However,with the increasing complexity of the tracking environment,there are some uncertain factors such as outliers interference and clutter in the measurement data,which leads to the serious decline of the estimation performance of the existing tracking technology.Therefore,how to design target tracking algorithms suitable for uncertain measurement environment to improve the accuracy of target state estimation is one of the urgent problems to be solved.The Bayesian filtering is one of the main forms of target tracking algorithm.The filter needs to build an accurate state space model for the target and realize state estimation through the Bayesian formula.Due to the existence of uncertain measurements,the corresponding state space model has the characteristics of multi-level and strong coupling,which makes it difficult to obtain the analytical solution of the target state.The variational Bayesian is an effective approximate solution method in the Bayesian framework.The algorithm decouples the estimated parameter set through the mean field theory,which effectively reduces the difficulty of calculating the approximate solution of the estimated parameters.In addition,the variational Bayesian obtains the approximate solution by minimizing the Kullback-Leibler divergence criterion,which theoretically ensures that the approximate solution can continuously approach the real value in the iterative process.In view of this,in this dissertation,the target tracking problems under uncertain measurement are studied based on variational Bayesian.The corresponding robust filtering algorithms are designed to improve the target tracking accuracy in the environment of non-Gaussian measurement noise and clutter,and the simulation experiments in typical application scenarios are carried out to verify the effectiveness of the proposed algorithms.The main work of the dissertation is as follows:1.Aiming at the problem of non-Gaussian characteristics of measurement noise caused by outliers interference,a Gaussian-Pearson type Ⅶ mixture distribution based model is constructed to fit the non-Gaussian measurement noise.The type of measurement noise is marked by the judgment factor which follows the Bernoulli-beta distribution,and the statistical characteristics of different types of measurement noise are further described in detail.Then,a Gaussian-Pearson type Ⅶ mixture distribution based robust Kalman filter is designed under the variational Bayesian framework to realize the joint estimation of measurement noise type,outlier noise covariance and target state,and effectively improve the target tracking accuracy under non-Gaussian measurement noise.2.Aiming at the problem of dynamic model mismatch caused by target maneuver uncertainty,the unknown input vector is introduced into the original state transition equation to compensate the dynamic model.At the same time,considering the problem that it is difficult to obtain the normal measurement noise covariance,which leads to the decline of measurement noise fitting,the Gaussian-inverse Wishart mixture distribution is used to model the non-Gaussian measurement noise.The input vector and the covariance of the measurement noise is estimated by using the variational Bayesian,which improves the target tracking accuracy under the dynamic model mismatch and non-Gaussian measurement noise.In addition,considering the non-zero mean characteristics of non-Gaussian measurement noise,a generalized hyperbolic unbiased student t-distribution based variational Bayesian Kalman filter with fading factor is proposed.The filter fits the non-zero mean non-Gaussian measurement noise through the generalized hyperbolic unbiased student t-distribution,and the fading factor is calculated by variational Bayesian to update the target state prediction covariance to achieve the model correction.3.Aiming at the low computational efficiency problem of probabilistic data association filter when the number of clutter is large,a novel probabilistic data association filter is designed under the variational Bayesian framework.Different from the traditional method,the algorithm realizes data fusion based on the weight Kullback-Leibler average,which avoids the calculation of cross covariance in the fusion process,and effectively improves the computational efficiency of the algorithm.In addition,for the condition that the unknown detection probability in clutter environment and the non-Gaussian process noise caused by violent maneuvering of targets,an adaptive probabilistic data association filter is designed in the variational Bayesian framework.Firstly,the detection probability and process noise are modeled by beta distribution and Pearson type Ⅶ distribution respectively.Then,to ensure the conjugation of parameters,the previous state is introduced into the parameter set and a new parameter estimation strategy is designed.Finally,the joint estimation of detection probability,process noise covariance and state is realized in the variational Bayesian framework,which effectively improves the target tracking accuracy under unknown detection probability and non-Gaussian process noise.4.Based on the multi-model system,the maneuvering target tracking under uncertain measurement is studied.Under the variational Bayesian framework,the interactive multimodel filters under non-Gaussian measurement noise and clutter are designed respectively.In addition,aiming at the problem that it is difficult to obtain the Markov transfor matrix in the traditional interactive multi-model filter,this problem is transformed into the inaccurate input of the sub-model,and a variational Bayesian interactive multi-model filter based on input correction is designed under the variational Bayesian framework to further improve the estimation accuracy of sub-model weight and target tracking.