Lightweight Marginalized Particle Filtering With Enhanced Consistency for Terrain Referenced Navigation

This article proposes a computationally lightweight marginalized particle filtering (MPF) algorithm with improved filter consistency. For mixed linear/nonlinear state-space models such as terrain referenced navigation (TRN) models, an MPF can improve the convergence and efficiency of the filters by separately applying a Kalman filter (KF) and a particle filter (PF) to each state space rather than using only one type of filter. However, an MPF still has a high degree of complexity because it requires the same number of KFs as the number of particles in the PF. To address this issue, our method simplifies the MPF through Gaussian approximation to use only one KF and improves the filter consistency compared to other MPF simplification methods. The proposed method further reduces the complexity by considering the dynamic model of a TRN system. This article presents a complexity analysis between the original MPF and other MPF simplification methods and numerical experiments for a range/bearing example and a TRN system. The results show that the proposed algorithm has better filter consistency than the existing simplification algorithms and can achieve performance comparable to an MPF with less computation.

Automated summary

This article proposes a computationally lightweight marginalized particle filtering (MPF) algorithm with improved filter consistency. The proposed algorithm improves the convergence and efficiency of the filters by separately applying a Kalman filter (KF) and a particle filter (PF) to each state space. By simplifying the MPF through Gaussian approximation and considering the dynamic model of a TRN system, the complexity of the algorithm is further reduced while maintaining filter consistency.