Localisation robuste d'un véhicule en environnement urbain à partir d'un système de stéréo-vision

Nicolas Simond 1
1 ICARE - Instrumentation, control and architecture of advanced robots
CRISAM - Inria Sophia Antipolis - Méditerranée
Abstract : The thesis deals with mobile robots in the specific context of urban environment. We want to extract the maximum of data on the motion of an uncalibrated stereo-rig embedded in a vehicle with in a structured environment. The aim of this study is to estimate the motion of a vehicle when the localization is impossible with a satellite position system due to the occlusion of the free space.

A polyhedrical model of the streets, which is the considered environment in this study, allows the segmentation of three main planes : the road and the vertical façades which represent the frontier of the navigation area. We assume that the white lanes which split the differents ways and the kerbs which delimit the road are all parallel in the scene, at least on the foreground. Therefore the detection of these induced edges in image is easy because they all converge in a dominant vanishing point.

The lighting conditions and the dynamical obstacles which occlude the vision field make the tracking task of the segmented planes difficult. The use of a stereo-vision system allows the segmentation in image of the coplanar area of the scene assuming that each plane induces a homography between both images of the stereo-rig. The vehicle motion is assumed to be plane, the pan and tilt rotation motions have low contributions on the computed homographies. We hence possess a reliable tool to segment the main planes which structure the environment.

The motion of the cameras between two poses are obtained with the homography induced by the relative motion of one plane. The homography computation requires the matching of features (lines, points) detected in the coplanar regions of images. The precision and the reliability of the result highly depends on the spatial distribution of the matched features.

The introduction of the notion of super-homography improves the robustness of the homography computation by computing in one shot all the homographies induced by the same plane in several images. The redundancy of the constraints imposed by the relative motion of the matched coplanar features increases the constraints of the plane projection in each image. The super-homography has two advantages: it allows the detection of mismatched features and supplies coordinates with a sub-pixellic precision. When the calibration of the stereo-rig is available, the extraction of the homography induced by the relative motion of a camera from the super-homography supplies the trajectory of the vehicle.
Document type :
Theses
Liste complète des métadonnées

https://tel.archives-ouvertes.fr/tel-00009694
Contributor : Simond Nicolas <>
Submitted on : Wednesday, July 6, 2005 - 3:26:45 PM
Last modification on : Saturday, January 27, 2018 - 1:31:01 AM
Document(s) archivé(s) le : Friday, April 2, 2010 - 9:56:48 PM

Identifiers

  • HAL Id : tel-00009694, version 1

Collections

Citation

Nicolas Simond. Localisation robuste d'un véhicule en environnement urbain à partir d'un système de stéréo-vision. Automatique / Robotique. Université Nice Sophia Antipolis, 2005. Français. ⟨tel-00009694⟩

Share

Metrics

Record views

410

Files downloads

486