Skip to Main content Skip to Navigation

Localisation et modélisation tridimensionnelles par approximations successives du modèle perspectif de caméra

Stéphane Christy 1
1 MOVI - Modeling, localization, recognition and interpretation in computer vision
GRAVIR - IMAG - Laboratoire d'informatique GRAphique, VIsion et Robotique de Grenoble, Inria Grenoble - Rhône-Alpes, CNRS - Centre National de la Recherche Scientifique : FR71
Abstract : In this report, we propose a generic algorithm to compute objet pose and reconstruction, with a perspective camera model. Given one image and a 3D model of the scene, object pose consists in recovering the position and the orientation of the camera with respect to the camera. We successively study the case of 2D to 3D point correspondences, and the case of line correspondences. The method consists in iteratively improving the pose computed with an affine camera model (weak perspective or paraperspective) to converge, at the limit, to the pose estimation computed with a perspective camera model. We analyze mathematical and geometric relationships between weak perspective, paraperspective and perspective camera models. We introduce a simple way to take into account the orthogonality constraint associated with the rotation matrix. We analyze the sensitivity to camera calibration errors and we define the optimal experimental setup with respect to imprecise camera calibration. We study its convergence based on numerical and experimental considerations, and we test its efficiency with both synthetic and real data. In a second time, we extend the previous object pose algorithms for Euclidean reconstruction from a sequences of images, by using a perspective camera model. The proposed method converges in a few iterations, is computationally efficient, and does not suffer from the non linear nature of the problem. With respect to factorization and/or affine-invariant methods, this method solves for the sign (reversal) ambiguity in a very simple way and provides much more accurate reconstruction results. We give a detailed account of the method and compare its complexity with respect to a non linear minimization method. Then, we present a second approach for recovering Euclidean reconstruction, with an uncalibrated affine camera mounted onto a robot arm. We show how using Euclidean information given by the robot motion. We also explain how obtaining camera calibration and hand-eye calibration. In order to use these algorithms for reconstruction from a practical point of view, we present a method to do the tracking of characteristic points along a sequence of images. Moreover, we also present a method to obtain a subpixel accuracy of the image point coordinates for a low computation cost.
Complete list of metadata

Cited literature [203 references]  Display  Hide  Download
Contributor : Thèses Imag <>
Submitted on : Thursday, February 19, 2004 - 2:24:24 PM
Last modification on : Monday, December 28, 2020 - 3:44:02 PM
Long-term archiving on: : Friday, April 2, 2010 - 8:32:16 PM


  • HAL Id : tel-00004885, version 1




Stéphane Christy. Localisation et modélisation tridimensionnelles par approximations successives du modèle perspectif de caméra. Interface homme-machine [cs.HC]. Institut National Polytechnique de Grenoble - INPG, 1998. Français. ⟨tel-00004885⟩



Record views


Files downloads