Skip to Main Content
The feasibility of a novel transformation, known as unscented transformation, which is designed to propagate information in the form of mean vector and covariance matrix through a non-linear process, is explored for underwater applications. The unscented transformation coupled with certain parts of the classic Kalman filter, provides a more accurate method than the EKF for nonlinear state estimation. Using bearings only measurements, Unscented Kalman filter algorithm estimates target motion parameters and detects target manoeuvre, using zero mean chi-square distributed random sequence residuals, in sliding window format. During the period of target manoeuvring, the covariance of the process noise is sufficiently increased in such away that, the disturbances in the solution is less. When target manoeuvre is completed, the covariance of process noise is lowered. In seawater, targets move at different speeds and will be at different ranges. It is observed that this algorithm is able to track all types of targets with encouraging convergence time. The performance of this algorithm is evaluated in Monte Carlo simulation and results are shown for various typical geometries.