Hierarchical particle filter for bearings-only tracking

We address here the classical bearings-only tracking problem (BOT) for a single target, an issue that belongs to the general class of nonlinear filtering problems. Recently, algorithm-based sequential Monte-Carlo methods (particle filtering) have been proposed. However, Fearnhead has observed that in practice this algorithm diverges. This problem is investigated further here. We show that this phenomenon is due to the unobservability of the distance between the observer and the target. We propose a new algorithm named hierarchical particle filter which takes into account this aspect of the BOT. We demonstrate that this novel filter architecture largely overperforms the classical one. Moreover, these results are confirmed when considering highly maneuvering target scenarios. Finally, we propose a general architecture based on Monte-Carlo methods for filtering initialization, able to accommodate poor prior and complex constraints.