Skip to Main content Skip to Navigation
Theses

Traitement automatique de vidéos en LSF Modélisation et exploitation des contraintes phonologiques du mouvement

Abstract : There are a lot of differences between sign languages and vocal languages. Among them, we can underline the simultaneity of several parameters, the important role of the face expression, the recurrent use of iconic gestures and the use of signing space to structure utterances. As a consequence, new methods have to be developed and adapted to those languages. At first, we detail a method based on a particle filter to estimate at any time, the position of the signer's head, hands, elbows and shoulders in a monoview video. This method has been adapted to the French Sign Language in order to make it more robust to occlusion, inversion of the signer's hands or disappearance of hands from the video frame. Then, we propose a classification of the motion patterns that are frequently involved in the sign production, thanks to the analysis of motion capture data. The parametric models associated to each sign pattern are used in the frame of automatic sign retrieval in a video from a filmed sign example. We finally include those models in two applications. The first one helps an user in creating sign pictures. The second one is dedicated to computer aided sign segmentation.
Complete list of metadatas

Cited literature [201 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00608768
Contributor : François Lefebvre-Albaret <>
Submitted on : Thursday, July 14, 2011 - 6:28:53 PM
Last modification on : Thursday, March 26, 2020 - 5:52:59 PM
Long-term archiving on: : Monday, November 12, 2012 - 11:05:23 AM

Identifiers

  • HAL Id : tel-00608768, version 1

Collections

Citation

François Lefebvre-Albaret. Traitement automatique de vidéos en LSF Modélisation et exploitation des contraintes phonologiques du mouvement. Sciences de l'ingénieur [physics]. Université Paul Sabatier - Toulouse III, 2010. Français. ⟨tel-00608768⟩

Share

Metrics

Record views

655

Files downloads

1660