A composite measurement based GLMB filter for multitarget tracking with multiple angle-only sensors

Modelling target birth is a significant issue that random finite set-based tracking filters must address. For angle-only sensors, because the measurements (azimuth and elevation angles) are nonlinear with the target full position (x-y-z Cartesian coordinates), the measurement-driven target birth model cannot be used directly. If take uniformly distributed Gaussian mixtures (GM) as the target birth model, it not only faces the tedious selection of parameters, but also poorly performs under low signal-to-noise ratio conditions. This paper proposes a composite measurement-based generalized labeled multi-Bernoulli (GLMB) filter for multiple angle-only sensors. First, the m-best S-dimensional assignment (SDA) algorithm is improved and applied to convert the angle measurements into a series of possible composite measurement sets. Then, the GLMB density is updated based on them. Since the composite measurement contains the target full position, it enables the proposed filter to use the measurement-driven target birth model accordingly. Moreover, by means of the improved m-best SDA algorithm, both the computational efficiency and the posterior completeness are guaranteed. The proposed filter is important for the applications where the target birth distribution is unknown. Its benefits are illustrated by the comparison with the existing multi-sensor GLMB filter equipped with uniformly distributed GM birth models.