Skip to Main content Skip to Navigation
Theses

Méthode de détection et de suivi multi-piétons multi-capteurs embarquée sur un véhicule routier: application à un environnement urbain

Abstract : At first my work consisted in developing an appropriate method to detect, then identify and track pedestrians in an outdoor environment from a single four-plane laser sensor. Pedestrian extraction and the merging of the four laser planes are both based on a nonparametric kernel method, also called "Parzen Windows". Initially, this estimator is used to approximate the likelihood function of the impact record in the laser image according to a pedestrian's geometrical characteristics. Secondly this estimator is used to calculate the likelihood that a pedestrian should be located within the four laser planes. Finally, to best characterize the complex trajectory of a pedestrian, the tracking process is based on the traditional particle filter. Unfortunately a pedestrian detection system which relies only on a laser sensor remains unsatisfactory as far as performance is concerned. Indeed, the inherent limitations of this sensor (no information about height, the outline or the color of the objects), as well as its sensitivity to such atmospheric conditions as rain or fog, make it necessary to resort to a multisensorial solution which allows to effectively combine the information provided by the laser and video sensors. This fusion methods is based on the development of a non-parametric method for data association, which allows to keep all the information contained in the measurements sent by the laser and video sensors. The performance of each proposed algorithm was characterized and reviewed, using real data obtained from numerous recording on board the LASMEA and Renault test vehicle ; Renault being the French vehicle manufacturer with whom we collaborate on our ANR LOVe project.
Document type :
Theses
Complete list of metadata

Cited literature [103 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00719262
Contributor : Camille Meyer <>
Submitted on : Thursday, July 19, 2012 - 2:08:32 PM
Last modification on : Thursday, January 11, 2018 - 6:17:20 AM
Long-term archiving on: : Friday, December 16, 2016 - 1:17:29 AM

Identifiers

  • HAL Id : tel-00719262, version 1

Citation

Samuel Gidel. Méthode de détection et de suivi multi-piétons multi-capteurs embarquée sur un véhicule routier: application à un environnement urbain. Automatique / Robotique. Université Blaise Pascal - Clermont-Ferrand II, 2010. Français. ⟨NNT : 2010CLF22028⟩. ⟨tel-00719262⟩

Share

Metrics

Record views

1147

Files downloads

7738