Target Tracking in the Presence of Intermittent Measurements via Motion Model Learning

When using a camera to estimate the pose of a moving target, measurements may only be available intermittently, due to feature tracking losses from occlusions or the limited field of view of the camera. Results spanning back to the Kalman filter have demonstrated the utility of using a predictor to...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on robotics 2018-06, Vol.34 (3), p.805-819
Hauptverfasser: Parikh, Anup, Kamalapurkar, Rushikesh, Dixon, Warren E.
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:When using a camera to estimate the pose of a moving target, measurements may only be available intermittently, due to feature tracking losses from occlusions or the limited field of view of the camera. Results spanning back to the Kalman filter have demonstrated the utility of using a predictor to update state estimates when measurements are not available, but target velocity measurements or a motion model must be known to implement a predictor for image-based pose estimation. In this paper, a novel estimator and predictor are developed to simultaneously learn a motion model, and estimate the pose, of a moving target from a moving camera. A stability analysis is provided to prove convergence of the state estimates and function approximation without requiring the restrictive persistent excitation condition. Two experiments illustrate the performance of the developed estimator and predictor. One experiment involves a stationary camera observing a mobile robot with sporadic feature tracking losses, and a second experiment involves a quadcopter moving between two mobile robots on a road network.
ISSN:1552-3098
1941-0468
DOI:10.1109/TRO.2018.2821169