Variational Extended Target Tracking With Dependent Measurements Using Random Matrix

The random matrix model provides a straightforward and low-complexity algorithm, making it a promising solution for extended target tracking and shape estimation. However, the random matrix methods in the literature generally operate under the assumption that the measurements are independently distributed, and based on this assumption, recursive filtering algorithms are designed to obtain the estimation of the state and the extent of the target. In practice, the measurements generated from the neighboring scattering centers on the surface of an extended target are often spatially correlated. Disregarding the dependence among the measurements may lead to a decline in the tracking performance. This article addresses both symmetric and asymmetric dependencies in the measurements for extended target tracking. Symmetric dependence is modeled using a Gaussian copula, while asymmetric dependence is captured with a skew-normal copula. Accordingly, the measurement likelihoods that account for correlation and skewness in the measurements are proposed. Based on the proposed models, we then derive the extended target tracking algorithms based on the variational Bayesian inference. The simulation results reveal that the proposed algorithm is effective, showing notable improvements compared to the current random matrix-based methods, particularly in scenarios where the measurement distributions exhibit high dependence.