Skip to Main content Skip to Navigation

Reconnaissance des gestes expressifs inspirée du modèle LMA pour une interaction naturelle homme-robot

Abstract : In this thesis, we deal with the problem of gesture recognition in a human-robot interaction context. New contributions are being made on this subject. Our system consists in recognizing human gestures based on a motion analysis method that describes movement in a precise way. As part of this study, a higher level module is integrated to recognize the emotions of the person through the movement of her body. Three approaches are carried out : the _rst deals with the recognition of dynamic gestures by applying the hidden Markov model (HMM) as a classi_cation method. A local motion descriptor is implemented based on a motion analysis method, called LMA (Laban Movement Analysis), which describes the movement of the person in its di_erent aspects. Our system is invariant to the initial positions and orientations of people. A sampling algorithm has been developed in order to reduce the size of our descriptor and also adapt the data to hidden Markov models. A contribution is made to HMMs to analyze the movement in two directions (its natural and opposite directions) and thus improve the classi_cation of similar gestures. Several experiments are done using public action databases, as well as our database composed of control gestures. In the second approach, an expressive gestures recognition system is set up to recognize the emotions of people through their gestures. A second contribution consists of the choice of a global motion descriptor based on the local characteristics proposed in the _rst approach to describe the entire gesture. The LMA E_ort component is quanti_ed to describe the expressiveness of the gesture with its four factors (space, time, weight and _ow). The classi_cation of expressive gestures is carried out with four well-known machine learning methods (random decision forests, multilayer perceptron, support vector machines : one-against-one and one-against-all. A comparative study is made between these 4 methods in order to choose the best one. The approach is validated with public databases and our database of expressive gestures. The third approach is a statistical study based on human perception to evaluate the recognition system as well as the proposed motion descriptor. This allows us to estimate the ability of our system to classify and analyze emotions as a human. In this part, two tasks are carried out with the two classi_ers (the RDF learning method that gave the best results in the second approach and the human classi_er) : the classi_cation of emotions and the 5 study of the importance of our motion features to discriminate each emotion.
Complete list of metadata
Contributor : Frédéric Davesne Connect in order to contact the contributor
Submitted on : Friday, January 25, 2019 - 11:58:46 AM
Last modification on : Saturday, May 1, 2021 - 3:48:46 AM
Long-term archiving on: : Friday, April 26, 2019 - 1:12:34 PM


Files produced by the author(s)


  • HAL Id : tel-01994111, version 1


Insaf Ajili. Reconnaissance des gestes expressifs inspirée du modèle LMA pour une interaction naturelle homme-robot. Apprentissage [cs.LG]. Université Paris-Saclay; Université d'Evry-Val-d'Essonne, 2018. Français. ⟨NNT : 2018SACLE037⟩. ⟨tel-01994111⟩



Record views


Files downloads