Types d'Actualités

Le jeudi 9 décembre à 14h, Mr Manuel ABBAS soutient sa thèse intitulée: "Detecting and analyzing physical activity in older adults using wearable sensors towards frailty trajectory assessment" devant le jury composé de:

  • Jérôme Boudy, Professeur, Institut Mines-Télécom Paris, Paris (rapporteur)
  •  Eric Campo, Professeur, Université Toulouse - Jean Jaurès, Toulouse (examinateur)
  • Jean-Baptiste Fasquel, Professeur, Université d'Angers, Angers (examinateur)
  • Pierre-Yves Guméry, Professeur, Université Grenoble Alpes, Grenoble (rapporteur)
  • Mohamad Khalil, Professeur, Lebanese University, Tripoli (examinateur)
  • Régine Le Bouquin Jeannès, Professeur, Université de Rennes 1, Rennes (directrice de thèse)
  • Dominique Somme, Professeur, CHU de Rennes - Service de Gériatrie, Rennes (examinateur)

Frailty is a geriatric syndrome characterized by physiological weakening that would affect 4 out of 5 people over 85 years old in France. In this context, our work consisted in proposing a wearable sensor-based fully automated system to monitor the activity of elderly with a view to analyze the frailty trajectory. This system consists of two layers, namely a machine learning-based human activity recognition (HAR) module to detect activities of daily living (ADLs), and a second module to analyze the ADLs. Regarding the first module, two HAR approaches were developed. The first one considers data acquired from an accelerometer, a gyroscope and a magnetometer, and is highly complex, based on sensing unit orientation, combining both handcrafted features and deep learning networks. The second one is an embeddable solution which exploits local temporal characteristics of acceleration signals exclusively. Once localized, the ADLs represented by windowed time series feed the second module. Here, two types of activity and health metrics are extracted from windowed data to assess the health conditions of elderly, namely global features which are computed over the course of a day, and local features which characterize the gait. A longitudinal study on data acquired under unsupervised conditions during the daily life of senior citizens (robust, pre-frail and frail) attested to the effectiveness and feasibility of our solution.