Research Of Underwater Target Tracking Technology Based On Nonlinear Filter

The ocean is essential for human beings to survive and necessary for sustainable development.As the development of the ocean continues to deepen,the role of science and technology in marine development is becoming more and more important.As an important part of ocean development,technology of underwater target tracking becomes a research hotspot that is paid more and more attention to.This paper mainly studies two key topics of underwater target tracking: filtering algorithm and target motion model.Firstly,the filtering algorithm in target tracking research is introduced in detail,including Kalman filtering algorithm,extended Kalman filter algorithm,unscented Kalman filter algorithm and cubature Kalman filter algorithm.A detailed introduction to the three motion models is presented,including the CV model,the CA model and the CT model.The estimation performance of three nonlinear filtering algorithms is analyzed and compared by simulation experiments.The simulation results show that the cubature Kalman filter algorithm has higher estimation accuracy than the extended Kalman filter algorithm and the unscented Kalman filter algorithm in the case of high dimensional state variables.Secondly,for the filter divergence problem of cubature Kalman filter caused by the covariance matrix losing positive definite,The square root cubature Kalman filter algorithm is introduced.Strong tracking cubature Kalman filter algorithm is introduced for the problem of filter divergence caused by inaccurate system model and state mutation.For the strong tracking cubature Kalman filter algorithm,the single fading factor can not correct the estimation of all state variables,so multiple fade factors square root cubature Kalman filter algorithm is proposed,which not only can solve the problem of filter divergence caused by the loss of positive definite of the covariance matrix but also can adjust the estimation of all state variables in the case of inaccurate models.The filtering results more stable and accurate.Simulation results also verify this.Finally,for the problem that target motion cannot be completely described by a single motion model,the Multiple Model algorithm is introduced.Especially the Interacting Multiple Model algorithm is introduced in detail.The Scalar-weight Interacting Multiple Model algorithm is also introduced.The model probability of the Interacting Multiple Model algorithm is calculated based on residual,and the model probability of the Scalar-weight Interacting Multiple Model algorithm is calculated based on covariance.If the model probability is calculated utilizing both residual and covariance,the resulting model probabilitywill be more accurate.Then the Fused Model Probability Interacting Multiple Model algorithm is proposed.The model probabilitys of Interacting Multiple Model algorithm and Scalar-weight Interacting Multiple Model algorithm are further fused to produce the new model probability,which is utilized to calculate the overall state estimation and the overall covariance.The simulation experiment shows that the Fused Model Probability Interacting Multiple Model algorithm has better estimation accuracy than the Interacting Multiple Model algorithm.