Suivi de paramètres de modèle géométrique à partir de séquences vidéo multi-vues

Abstract : Various applications, such as augmented reality, are based on the need to follow the movements or changes of real objects in video sequences. We present in this thesis a method for monitoring articulated subjects in videos. The proposed method is based on the use of a generic model and estimation of its parameters from CGI. In our problem, we have several calibrated cameras filming the subject as well as a representation of the subject as a parameterized geometrical model. With a renderer, we can generate synthetic images of the subject corresponding to each view. It is then to find the parameters that generate the most accurate images from the actual images, considered as reference images. This involves the introduction of a measure of multiview gap between synthetic and real image and the search for the optimum of this measure over model parameters. The gap defined measure is based on a comparison between dense images, particularly with a strong resistance to occlusions. The search for the optimum measurement is based on the principle of the simplex, adapted to our situation. For each time we seek the values of the corresponding parameters, and we spread them as initial values for the next moment. We tested our method on synthetic sequences to demonstrate its theoretical validity. We also studied the conditions affecting tracking accuracy. Tests on real sequences show the ability to work on concrete examples.
Liste complète des métadonnées

https://hal.archives-ouvertes.fr/tel-01110256
Contributor : Yannick Perret <>
Submitted on : Wednesday, January 28, 2015 - 12:02:06 PM
Last modification on : Monday, June 11, 2018 - 3:46:01 PM
Document(s) archivé(s) le : Wednesday, April 29, 2015 - 10:25:16 AM

Licence


Distributed under a Creative Commons Attribution 4.0 International License

Identifiers

  • HAL Id : tel-01110256, version 1

Collections

Citation

Yannick Perret. Suivi de paramètres de modèle géométrique à partir de séquences vidéo multi-vues. Informatique [cs]. Universite Claude Bernard Lyon 1, 2001. Français. ⟨NNT : 2001LYO10276⟩. ⟨tel-01110256⟩

Share

Metrics

Record views

89

Files downloads

101