Skip to Main content Skip to Navigation

Stéréovision embarquée sur véhicule : de l'auto-calibrage à la détection d'obstacles

Abstract : The increasing complexity of road environments and the will to improve driving safety may explain the large number of studies (investigations / research programs ) on driving assistance. Numerous systems (based on ultrasonic sensors, laser range, radar,...) have been set up to perceive the environment around a vehicle. All of them have to deal with the diversity of scenes observed, both in urban and road contexts. Therefore, one has to use sensors providing rich information that can be analysedin several functions and in any configuration. Vision seems to be a relevant choice. It is indeed an essential sensorial source for human's decision making process. Thus, the methods described in this thesis consist in proposing a complete and autonomous passive stereovision system for obstacles detection on a so-called "intelligent" vehicle, evolving in urban or road environments. The main issue about using a sterovision bench on board a robot or a vehicle concerns the calibration accuracy, which tend to vary with time. Indeed, such a system can only be reliable if the parameters describing the sensor's geometry can be correctly estimated, in spite of vibrations and possible shocks. First, an on-line recalibration method for an initially calibrated system will be presented. It consists in a dynamic correction of its parameters. Once calibration issue is tackled, the second one considered is obstacles detection in road environment. The methods described in this document propose a building of the longitudinal profile of the road so as to extract an obstacle model from it, which takes into account the variations in pitch (superelevation) and/or roll. This approach processes 3D data provided by stereovision. The last application concerns a perception system for parking assistance; a generic method for an incremental building of a 3D model of free space will be presented. That method assumes that the vehicle movement is estimated with a good precision. The 3D model obtained is th en analysed to detect a parking space and estimate its characteristics; the goal is to provide information to a motion planning system enabling an automatic achievement of the parking manoeuvring. Every method has been integrated in the RT-MAPS environment, and validated on-line on board one of the LAAS robots or with images sequences taken on board vehicles.
Complete list of metadata

Cited literature [10 references]  Display  Hide  Download
Contributor : Catherine Martineau <>
Submitted on : Thursday, March 23, 2006 - 11:29:18 AM
Last modification on : Friday, October 23, 2020 - 4:33:45 PM
Long-term archiving on: : Saturday, April 3, 2010 - 10:02:55 PM


  • HAL Id : tel-00012017, version 1


Vincent Lemonde. Stéréovision embarquée sur véhicule : de l'auto-calibrage à la détection d'obstacles. Réseaux et télécommunications [cs.NI]. INSA de Toulouse, 2005. Français. ⟨tel-00012017⟩



Record views


Files downloads