| Summary: | In this paper, we propose a method for optimal stochastic sensor control, where the goal is to minimise the estimation error in multi-object tracking scenarios. Our approach is based on an information theoretic divergence measure between labelled random finite set densities. The multi-target posteriors are generalised labelled multi-Bernoulli (GLMB) densities, which do not permit closed form solutions for traditional information divergence measures such as Kullback-Leibler or Rényi. However, we demonstrate that the Cauchy-Schwarz divergence admits a closed form solution for GLMB densities, thus it can be used as a tractable objective function for multi-target sensor control. This is demonstrated with an application to sensor trajectory optimisation for bearings-only multi-target tracking.
|
|---|