Improving monocular visual odometry localization with wheel and inertial measurements
Résumé
Global Navigation Satellite System (GNSS) is a popular localization system for autonomous vehicles, but an onboard localization system is often required to take over when the signal degrades or is absent, for example in difficult conditions (tunnel, building, forest etc.). Among the existing possibilities, visual odometry is often selected. It does not need any prior knowledge, and is accurate enough if used for a limited period of time. However, it has also several drawbacks: low frequency (due to important processing time) and inconsistency (due to low illumination, glare conditions or moving objects). Moreover, when using a monocular camera, the localization is obtained up to a scale factor. These drawbacks can be solved using other onboard sensors. The most common sensors for autonomous vehicles are inertial measurement units (IMU), wheel encoders, or Lidars. In this paper, a comparison of visual, inertial and wheel odometries (VO, IO and WO respectively) is proposed. The goal is to precisely evaluate their performances on the estimation of the 2D pose (position and heading) of an autonomous platform. With this information, our goal is to increase frequency, and correct inconsistencies of our VO localization. First results correcting the scale and improving the accuracy will be demonstrated using a ground vehicle in outdoor environments.
Domaines
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |