Skip to Main content Skip to Navigation

Renforcement des méthodes de prise de décision par des a priori pour la mesure automatique de l'activité physique des personnes

Abstract : Advances in technology have led to the miniaturization of motion sensors, facilitating their use in small comfortable wearable devices. Such devices are of great interest in the biomedical field especially for applications aimed at estimating the daily physical activity of people. In this thesis, we propose signal processing algorithms allowing better interpretation of sensors measures and thus their mapping to different activities. Our approach is based on that activities have strong temporal dependencies. We propose a recognition system that models the activity sequence by a Markov chain. Our system relies on parametric and non-parametric classification methods. The soft output of classifiers permits the construction of confidence measures in the activities. These measures are later used as input to a Viterbi algorithm that gives the final estimation of the activity sequence. We validate our algorithms using a database containing 48 subjects, each of whom having carried out activities for more than 90 minutes. Moreover, this thesis aims at providing practical answers to challenges concerning the development of an activity recognition system. First of all, we wonder about the optimal sensor placement on the body, and about the number of sensors needed for a reliable estimation of activities. We also approach the problem of selecting relevant features for classifiers. Another crucial issue concerns the estimation of the sensor’s orientation on the body : this involves the problem of sensor calibration. Finally, we provide a “real-time” implementation of our system, and collect a database under realistic conditions to validate our implemented real-time demostrator.
Keywords : Actigraphy
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Monday, May 11, 2015 - 5:48:06 PM
Last modification on : Friday, July 31, 2020 - 10:44:08 AM
Long-term archiving on: : Monday, September 14, 2015 - 10:35:43 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01150746, version 1



Abbas Ataya. Renforcement des méthodes de prise de décision par des a priori pour la mesure automatique de l'activité physique des personnes. Traitement du signal et de l'image [eess.SP]. Télécom ParisTech, 2013. Français. ⟨NNT : 2013ENST0071⟩. ⟨tel-01150746⟩



Record views


Files downloads