Skip to Main content Skip to Navigation
Theses

Fusion de données multi-capteurs à l'aide d'un réseau bayésien pour l'estimation d'état d'un véhicule

Cherif Smaili 1
1 MAIA - Autonomous intelligent machine
INRIA Lorraine, LORIA - Laboratoire Lorrain de Recherche en Informatique et ses Applications
Abstract : This thesis presents a multi-sensor fusion strategy for a novel road-matching based on Bayesian Network. %Localization sensors are generally imperfect and provide only uncertain information. A localization method that seeks to be reliable must use a methodology able to handle uncertain and ambiguous information. An outdoor vehicle equipped with a GPS receiver is faced with a pose-tracking problem since the probability of not having a GPS fix for a long distance is very low, even though the GPS system can be intermittent in urban environments for example. Therefore, the work reported in this thesis concerning the mono-vehicle localization is motivated by the need to build a robust pose tracking system based on road-matching module.\\ Road-matching is a technique that attempts to locate an estimated vehicle position on road network. Many road-matching algorithms have been developed and widely incorporated into GPS/DR vehicle navigation systems for both commercial and experimental ITS applications. However, simply combining GPS and DR cannot provide an accurate vehicle positioning system. In that case, we use the digital map as an observation to improve the reliability of road-matching. \\ In many cases, when a vehicle is in front ambiguous situations (close roads, junction roads...), we try to identify the most likely segment. Inevitably, this identification is closely related to the accuracy of sensors and therefore the identification of the most likely segment remains a real problem. To resolve these ambiguous situations, this work presents a multi-sensor fusion and multi-modal estimation realized using Bayesian Network. The strategy presented in this work doe not keep only the most likely segment. When approaching an intersection, several roads can be good candidates for this reason we manage several hypotheses until the situation becomes unambiguous.\\ Multi-vehicle localisation method presented in this work can be seen like an extension of the method presented above. In the sense that we have duplicate the Bayesian Network used to fuse measurements sensors to localize one vehicle for several ones. Than, we have added vehicles inter-connexion to represent finally the platoon of vehicles in the context of Bayesian Network. This kind of platoon representation for vehicles poses estimations permits additionally to implement control law based on near-to-near approach, which can only be seen as a first step in platoon control design. \\ The use of Extended Kalman Filter has been extensively used for localizing such a vehicle from multi-sensors data fusion. The non-linearity of the vehicle kinematic model gave raise to several extensions of classical Kalman filters such as Extended Kalman Filter and Unscented Kalman Filter. In order to avoid the linearization process, we propose to take advantage of the possibility of transforming the kinematic system in chained form using an appropriate feedback transformation. With the linear form (chained form), an exact inference computation on the Bayesian network can be performed. Real data are used (ABS sensors, a differential GPS receiver and an accurate digital roadmap) to illustrate the performance of our approach.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00551833
Contributor : François Charpillet <>
Submitted on : Tuesday, January 4, 2011 - 4:41:34 PM
Last modification on : Monday, April 16, 2018 - 10:42:05 AM
Long-term archiving on: : Tuesday, April 5, 2011 - 3:05:22 AM

Identifiers

  • HAL Id : tel-00551833, version 1

Collections

Citation

Cherif Smaili. Fusion de données multi-capteurs à l'aide d'un réseau bayésien pour l'estimation d'état d'un véhicule. Informatique [cs]. Université Nancy II, 2010. Français. ⟨tel-00551833⟩

Share

Metrics

Record views

5826

Files downloads

1069