Impact of landmark parametrization on monocular EKF-SLAM with points and lines - Archive ouverte HAL Accéder directement au contenu
Pré-Publication, Document De Travail Année : 2010

Impact of landmark parametrization on monocular EKF-SLAM with points and lines

Résumé

This paper aims at providing answers to a variety of questions regarding undelayed initialization of point- and line-landmarks in monocular EKF-SLAM. It does so by stating the problem and the necessary conditions for a successful solution, and by offering a compendium of landmark parametrizations, including three different methods for points and five other methods for straight lines. Some of these methods are already well-known, other are presented here for the first time. The discourse of the text is chosen to highlight the similarities and the differences between them, and how all these parametrizations can be seen as incremental variations of ones with respect to others. We show unified approaches for points and lines and explain how they could be extended to conics, and why they cannot be easily generalized to planes. We present a new linearity measure to predict how well a parametrization will perform in an EKF framework. To confirm these predictions, we benchmark all the methods by running Monte-Carlo tests on all algorithms, using the same data and comparing their performances on filter consistency. Finally, some results with real imagery are presented as a means to further analyze their possibilities in view of real implementations.
Fichier principal
Vignette du fichier
solaTRO10.pdf (772.34 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-00451778 , version 1 (31-01-2010)
hal-00451778 , version 2 (05-06-2010)
hal-00451778 , version 3 (01-01-2011)
hal-00451778 , version 4 (17-09-2011)

Identifiants

  • HAL Id : hal-00451778 , version 1

Citer

Joan Solà, Teresa Vidal-Calleja, Javier Civera. Impact of landmark parametrization on monocular EKF-SLAM with points and lines. 2010. ⟨hal-00451778v1⟩
414 Consultations
2144 Téléchargements

Partager

Gmail Facebook X LinkedIn More