A Robust Descriptor for Tracking Vertical Lines in Omnidirectional Images and its Use in Mobile Robotics

Abstract : In this paper we introduce a robust descriptor for matching vertical lines among two or more images from an omnidirectional camera. Furthermore, in order to make such a descriptor usable in the framework of indoor mobile robotics, this paper introduces a new simple strategy to extrinsically self-calibrate the omnidirectional sensor with the odometry reference system. In the first part of this paper we describe how to build the feature descriptor.We show that the descriptor is very distinctive and is invariant to rotation and slight changes in illumination. The robustness of the descriptor is validated through real experiments on a wheeled robot. The second part of the paper is devoted to the extrinsic self-calibration of the camera with the odometry reference system. We show that by implementing an extended Kalman filter that fuses the information from the visual features with the odometry, it is possible to extrinsically and automatically calibrate the camera while the robot is moving. In particular, it is theoretically shown that only one feature suffices to perform the calibration. Experimental results validate the theoretical contributions.
Document type :
Journal articles
Liste complète des métadonnées

Cited literature [37 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-00428670
Contributor : Agostino Martinelli <>
Submitted on : Thursday, October 29, 2009 - 5:29:50 PM
Last modification on : Tuesday, April 2, 2019 - 2:03:11 AM
Document(s) archivé(s) le : Thursday, June 17, 2010 - 6:39:35 PM

File

scaramuzza_ijrr2.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-00428670, version 1

Collections

Citation

Davide Scaramuzza, Agostino Martinelli, Roland Siegwart. A Robust Descriptor for Tracking Vertical Lines in Omnidirectional Images and its Use in Mobile Robotics. International Journal on Robotics Research, 2009, pp.000. ⟨hal-00428670⟩

Share

Metrics

Record views

449

Files downloads

380