Skip to Main content Skip to Navigation

Classification and Characterization of Emotional Body Expression in Daily Actions

Abstract : The work conducted in this thesis can be summarized into four main steps. Firstly, we proposed a multi-level body movement notation system that allows the description of expressive body movement across various body actions. Secondly, we collected a new database of emotional body expression in daily actions. This database constitutes a large repository of bodily expression of emotions including the expression of 8 emotions in 7 actions, combining video and motion capture recordings and resulting in more than 8000 sequences of expressive behaviors. Thirdly, we explored the classification of emotions based on our multi-level body movement notation system. Random Forest approach is used for this purpose. The advantage of using Random Forest approach in our work is double-fold : 1) reliability of the classification model and 2) possibility to select a subset of relevant features based on their relevance measures. We also compared the automatic classification of emotions with human perception of emotions expressed in different actions. Finally, we extracted the most relevant features that capture the expressive content of the motion based on the relevance measure of features returned by the Random Forest model. We used this subset of features to explore the characterization of emotional body expression across different actions. A Decision Tree model was used for this purpose.
Complete list of metadata
Contributor : Nesrine Fourati <>
Submitted on : Friday, March 4, 2016 - 12:40:26 PM
Last modification on : Friday, July 31, 2020 - 10:44:08 AM
Long-term archiving on: : Sunday, June 5, 2016 - 10:31:30 AM


Distributed under a Creative Commons Attribution - NonCommercial - ShareAlike 4.0 International License


  • HAL Id : tel-01282785, version 1



Nesrine Fourati. Classification and Characterization of Emotional Body Expression in Daily Actions. Computer Science [cs]. Telecom ParisTech, 2015. English. ⟨tel-01282785⟩



Record views


Files downloads