Observability Criteria and Null-Measurement Kalman Filter for Vision-Aided Inertial Navigation

Abstract

This work presents a linear time-varying approach to real-time vision-aided inertial navigation. An exact measurement model is provided, which linearly depends on the state and is therefore suitable for Kalman filtering. Sufficient observability and nonobservability criteria are derived, which can be used to identify observable trajectories. Moreover, the paper proposes a filter able to determine the absolute scale of the camera motion without any need for estimating three-dimensional feature positions or a priori knowledge of the absolute attitude. The effectiveness of the filter on low-cost hardware is demonstrated in flight tests with a small fixed-wing unmanned aerial vehicle.