A Gaussian mixture motion model and contact fusion applied to the Metron data set

In active sonar, low measurement rates of up to several minutes are commonly used. Thus, strong manoeuvres may occur between two measurements. To assure high-quality tracking, target dynamics have to be modelled precisely. In this paper, a variant of the Nearly Constant Velocity model is presented with deviations modelled according to the assumption that for submarine manoeuvres a change in heading is more likely than a change in speed. The proposed motion model uses Gaussian Mixtures to approximate the non-Gaussian distribution after the Kalman Filter prediction step. It is realised in the framework of multi-hypothesis tracking and applied to the Metron data set which simulates sonar contacts detected by 25 distributed sonobuoys. A high amount of low-quality input data is provided to the tracker. Thus, this paper additionally proposes a contact-fusion approach to display the measurements of a single time step and of multiple sensors into Cartesian coordinates.