Metric Visual-Inertial Ego-Motion Estimation Using Single Optical Flow Feature
Résumé
This paper introduces a state estimation framework that allows estimating the attitude, full metric speed and the orthogonal metric distance of an IMU-camera system with respect to a plane. The filter relies only on a single optical flow
feature as well as gyroscope and accelerometer measurements. The underlying assumption is that the observed visual feature lies on a static plane. The orientation of the observed plane is not required to be known a priori and is also estimated at runtime. The estimation framework fuses visual and inertial measurements in an Unscented Kalman Filter (UKF). The theoretical limitations of the UKF are investigated using a nonlinear observability analysis based on Lie-derivatives. Experiments in simulation using realistic sensor noise values
successfully demonstrate the performance of the filter as well as validate the findings of the observability analysis. It is shown that the state estimate is converging correctly, even in presence of substantial initial state errors. To the authors’ knowledge, this paper documents for the first time the estimation of the
heading and metric distance to a wall with no range- or bearing sensors, relying solely on optical flow as the only exteroceptive sensing modality. This minimal sensor set, that is both lightweight and low-cost, renders the framework an appealing choice for the use as a navigation system on a wide range of robotic
platforms, such as ground- or flying robots.
Domaines
Sciences de l'ingénieur [physics]
Origine : Fichiers produits par l'(les) auteur(s)