Adaptive measurement conversion for underwater target tracking with unknown non-Gaussian noise

Aiming at the non-Gaussian polar-Cartesian underwater target tracking problem with unknown measurement outliers, an iterative converted measurement Student’s t filter based on the variational Bayesian method (VBICMSTF) is proposed. Range and azimuth estimations of active sonar target are taken as nonlinear measurement based on polar coordinates, and the pseudo-linear measurement after unbiased conversion is modeled approximately using student’s t distribution. Then, the posterior distributions of the pseudo-linear measurement scale array and the target state are iteratively updated by the variational Bayesian method. During the iteration process, the updated target position is used to correct the prior calculation of the second moment of measurement conversion, forming a prior-posterior update loop. Simulation and lake experimental results show that the VBICMSTF reduces the tracking error by more than 25% compared with the pseudo-linear Student’s t distribution variational Bayesian algorithm, and maintains the consistency of filtering in the strong nonlinear tracking scene with unknown non-Gaussian measurement noise.