A long simulation scenario for evaluation of multi-target tracking methods

Abstract

Algorithms for multi-target tracking are typically evaluated by means of simulations. Typically, the simulated scenarios are fairly short and of very simple geometry. In contrast, this paper describes a scenario which was deliberately and explicitly designed to exhibit a large challenging behavior for several targets moving in close formation for about 23 minutes.The scenario is inspired by experiments with autonomous surface vehicles (ASVs). We use the scenario to study the performance differences between tracking methods in the Poisson Multi-Bernoulli Mixture (PMBM) family. A multiple models approach outperforms the single model approach, but is to a large extent outperformed by the multiple hypothesis approach.