Skip to Main content Skip to Navigation

Reconnaissance d’expressions corporelles dans des mouvements de personnes en vue de la synthèse de style

Abstract : The theme of my thesis concerns the recognition and synthesis of facial and body expressions. Our problem is to study, understand and extract the elements that translate a person's emotional state from the expressions of his face and body, in order to recognize and also to synthesize the emotion or style in a gesture. This dual objective of recognition and synthesis will make it possible to generalize new modes of interaction in applications such as video games, human-machine interaction, etc. Indeed, these applications can be enriched with scenarios that could be adapted to the emotional state of the user. To answer this problem, the crucial point is to understand "where" the expression information is located in an action. Indeed, in a few seconds, a human knows how to characterize an expression he sees while recognition algorithms are still far from it, especially for body expressions where, to our knowledge, little work has been done compared to the recognition of facial expressions. Concerning the recognition of expressions, our objective is, first of all, to propose features capable of recognizing the expression carried by a movement. To do this, the main problem is to separate the movement achieved from the perceived expression. Concerning the recognition of facial expressions, we are interested in the societal problem of parental protection. To do this, it is necessary to understand and know how to recognize facial expressions of children. To solve this problem, we have built and proposed a new database to help the computer vision community understand the specificities of facial expressions of children's faces. Secondly, we also hope that the various descriptors proposed in recognition of body expressions can be used in the field of animation synthesis. Indeed, in the field of animation, the creation of an action that conveys an emotion or a style requires a lot of work, know-how and time for an animator to propose stylized animations. For example, in a video game, to create such animations, it is often necessary to have a huge database of movements including each style for each virtual character. To have such a database, one of the following two methods is often used. The first consists in capturing all the movements made by different actors playing different styles. In the second, it is the graphic designer who must create the various animations by hand using animation software. Our objective in this context is to use the descriptors quantifying the expression detected in recognition of facial and body expressions in order to develop tools capable of changing / editing the style or expression of an animation. These tools will assist and facilitate the work of graphic designers by allowing them to quickly synthesize a stylized "primal animation". These stylized animations can be refined in post-processing by adding an artistic touch to the generated animation.
Complete list of metadata

Cited literature [155 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Wednesday, May 20, 2020 - 10:06:10 AM
Last modification on : Tuesday, June 1, 2021 - 2:08:10 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02613404, version 1


Arthur Crenn. Reconnaissance d’expressions corporelles dans des mouvements de personnes en vue de la synthèse de style. Vision par ordinateur et reconnaissance de formes [cs.CV]. Université de Lyon, 2019. Français. ⟨NNT : 2019LYSE1341⟩. ⟨tel-02613404⟩



Record views


Files downloads