MINES ParisTech CAS - Centre automatique et systèmes

Fusion of inertial and visual: a geometrical observer-based approach

Authors: S. Bonnabel, P. Rouchon Intelligent Systems and Automation: 2nd Mediterranean Conference on Intelligent Systems and Automation (CISA'09) Vol 1107 no 1 pp. 54-58 DOI: 10.1063/1.3106512
The problem of combination between inertial sensors and CCD cameras is of paramount importance in various applications in robotics and autonomous navigation. In this paper we develop a totally geometric model for analysis of this problem, independently from a camera model and from the structure of the scene (landmarks etc.). This formulation can be used for data fusion in several inertial navigation problems. The estimation is then decoupled from the structure of the scene. We use it in the particular case of the estimation of the gyroscopes bias and we build a nonlinear observer which is easy to compute, provides an estimation of the biais, filters the image, and is by construction very robust to noise.
Download PDF
BibTeX:
@Proceedings{,
author = {S. Bonnabel, P. Rouchon},
editor = {},
title = {Fusion of inertial and visual: a geometrical observer-based approach},
booktitle = {Intelligent Systems and Automation: 2nd Mediterranean Conference on Intelligent Systems and Automation (CISA’09)},
volume = {1107 (1)},
publisher = {American Institute of Physics},
address = {},
pages = {54-58},
year = {2009},
abstract = {The problem of combination between inertial sensors and CCD cameras is of paramount importance in various applications in robotics and autonomous navigation. In this paper we develop a totally geometric model for analysis of this problem, independently from a camera model and from the structure of the scene (landmarks etc.). This formulation can be used for data fusion in several inertial navigation problems. The estimation is then decoupled from the structure of the scene. We use it in the particular case of the estimation of the gyroscopes bias and we build a nonlinear observer which is easy to compute, provides an estimation of the biais, filters the image, and is by construction very robust to noise.},
keywords = {nonlinear systems, geometry, numerical analysis}}