Towards Consistent Visual-Inertial Navigation for Unmanned Aerial Vehicles using Depth Information

This paper proposes a new filtering-based depth enhanced visual inertial navigation system (DE-VINS) for quadrotor unmanned aerial vehicles (UAV). This filter addresses the drifting and degraded performance of a class of conventional monocular VINS filters at hovering conditions. A theoretical nonlinear observability analysis is performed to verify the filter design. The performance of the proposed DE-VINS is numerically evaluated using a Matlab simulator and then compared against the conventional VINS filter proposed in literature. The results show improved performance of the DE-VINS in terms of estimation accuracy and consistency at zero-velocity flight conditions.