Skip to Main content Skip to Navigation

Localisation d'un véhicule à l'aide d'un SLAM visuel contraint

Abstract : To ensure a global localization in urban environment, the majority of commercial solutions is based on Global Positioning Systems (GPS). While these systems offer sufficient accuracy in peri-urban or rural areas, their accuracy decreases significantly in cities because of the urban canyon (i.e. reflections of the GPS signal through the facades of buildings). To overcome this problem, vision based solutions such as the visual SLAM (Simultaneous Localization And Mapping) seem to be a promising alternative. In addition to the estimation of the six degrees of freedom of the mobile camera, such approach provides a 3D map of the observed scene. However, the localization provided by the visual SLAM is not geo-referenced and is often subject of drifts (e.g. poor estimate of the scale factor, accumulations errors). To address these limitations and to provide a solution easy to deploy, we studied the possibility of integrating to the SLAM algorithm additional information that could constrain the entire reconstruction. These data must then be inexpensive and available in dense urban and peri-urban areas. For these reasons, we chose to exploit the constraints provided by a standard GPS and those provided by models from Geographic Information Systems, more precisely, the 3D buildings models and the digital elevation models. The main contribution of this work lies in the integration of these constraints in the bundle adjustment (i.e. the optimization process of the SLAM algorithm). This is not trivial since combining constraints acting on the trajectory of the camera and the 3D reconstruction can lead to convergence problems, especially when the information used have various uncertainties and even outliers (e.g. specially GPS measurements). Different solutions Larnaout et al. (2012, 2013a,b,c) to combine these constraints simultaneously while limiting the problems of convergence have been developed. The proposed solutions have been evaluated on synthetic sequences and large scale real sequences recorded in normal traffic conditions. The results show that the accuracy achieved on the six degrees of freedom of the mobile camera is sufficient to ensure new service of aided navigation through Augmented Reality. In addition to the accuracy, our approaches have the advantage of being fast, inexpensive and easy to deploy (not requiring sophisticated equipment).
Document type :
Complete list of metadata

Cited literature [103 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Wednesday, July 23, 2014 - 10:27:10 AM
Last modification on : Tuesday, April 20, 2021 - 11:22:12 AM
Long-term archiving on: : Tuesday, April 11, 2017 - 4:59:03 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01038016, version 1


Dorra Larnaout. Localisation d'un véhicule à l'aide d'un SLAM visuel contraint. Autre. Université Blaise Pascal - Clermont-Ferrand II, 2014. Français. ⟨NNT : 2014CLF22454⟩. ⟨tel-01038016⟩



Record views


Files downloads