Particle filter and Gaussian-mixture filter efficiency evaluation for terrain-aided navigation

Terrain-aided navigation is a method relying on a digital terrain elevation database and radar-altimeter measurements and can be applied to manned or unmanned aircrafts. Associated with an inertial navigation system, terrain-aided navigation provides an accurate estimation of position. Since the aircraft state estimation implies non-linear filtering, the computational load of terrain-aided navigation algorithms is generally high. Hence, for real-time implementation, non-linear filters should be designed to achieve maximum performances with limited resources. In this work, we focus on particle filter and Gaussian-mixture filter which are two classical approaches to solve non-linear problems in a Bayesian framework. We describe the two algorithms and compare their performances on various terrain topographies. These simulations highlight that the Gaussian-mixture filter achieves better performances and reliability, in a situation where the filter design aims at reducing computational requirements.