Le réseau de capteurs porté est une technologie d’avenir prometteuse à multiple domaines d’application allant du médical à l’interface homme machine. Le projet BoWI a pour ambition d’évaluer la possibilité d’élaborer un réseau de capteurs utilisable au quotidien dans un large spectre d’applications et ergonomiquement acceptable pour le grand public. Cela induit la nécessité de concevoir un nœud de réseau ultra basse consommation pour à la fois convenir à une utilisation prolongée et sans encombrement pour le porteur. La solution retenue est de concevoir un nœud capable de travailler avec une énergie comparable à ce que l’état de l’art de la récolte d’énergie est capable de fournir. Une solution ASIC est privilégiée afin de tenir les contraintes d’intégration et de basse consommation. La conception de l’architecture dédiée a nécessité une étude préalable à plusieurs niveaux. Celle-ci comprend un état de l’art de la récolte d’énergie afin de fixer un objectif de budget énergie/puissance de notre système. Une étude des usages du système a été nécessaire notamment pour la reconnaissance postures afin de déterminer les cas d’applications types. Cette étude a conduit au développement d’algorithmes permettant de répondre aux applications choisies tout en s’assurant de la viabilité de leurs implantations. Le budget énergie fixé est un objectif de 100µW. Les applications choisies sont la reconnaissance de posture, la reconnaissance de geste et la capture de mouvement. Les solutions algorithmiques choisis sont une fusion de données de capteurs inertiels par Filtre de Kalman étendu (EKF) et l’ajout d’une classification par analyse en composante principale. La solution retenue pour obtenir des résultats d’implémentation est la synthèse de haut niveau qui permet un développement rapide. Les résultats de l’implantation matérielle sont dominés principalement par l’EKF. À la suite de l’étude, il apparait qu’il est possible avec une technologie 28nm d’atteindre les objectifs de budget énergie pour la partie algorithme. Une évaluation de la gestion haut niveau de tous les composants du nœud est également effectuée afin de donner une estimation plus précise des performances du système dans un cas d’application réel. Une contribution supplémentaire est obtenue avec l’ajout de la détection d’activité qui permet de prédire la charge de calcul nécessaire et d’adapter dynamiquement l’utilisation des ressources de traitement et des capteurs afin d’optimiser l’énergie en fonction de l’activité / Wireless Body Sensor Network (WBSN) is a promising technology that can be used in a lot of application domains from health care to Human Machine Interface (HMI). The BoWI project ambition is to evaluate and design a WBSN that can be used in various applications with daily usage and accessible to the public. This necessitates to design a ultra-low power node that reach a day of use without discomfort for the user. The elected solution is to design a node that operates with the power budget similar to what can be provided by the state of the art of the energy harvesting. An Application Specific Integrated Circuit (ASIC) solution is privileged in order to meet the integration and low power constraints. Designing the dedicated architecture required a preliminary study at several level which are: a state of the art of the energy harvesting in order to determine the objective of energy/power budget of our system, A study of the usage of the system to determine and select typical application cases. A study of the algorithms to address the selected applications while considering the implementation viability of the solutions. The power budget objective is set to 100µW. The application selected are the posture recognition, the gesture recognition and the motion capture. The algorithmic solution proposed are a data-fusion based on an Extended Kalman FIlter (EKF) with the addition of a classification using Principal Component Analysis (PCA). The implementation tool used to design the architecture is an High Level Synthesis (HLS) solution. Implementation results mainly focus on the EKF since this is by far the most power consuming digital part of the system. Using a 28nm technology the power budget objective can be reached for the algorithmic part. A study of the top level management of all components of the node is done in order to estimate performances of the system in real application case. This is possible using an activity detection which dynamically estimates the computing load required and then save a maximum of energy while the node is still.

Made available in DSpace on 2016-06-02T19:05:57Z (GMT). No. of bitstreams: 1 4452.pdf: 7806002 bytes, checksum: 03f3df43afda775ba6b11c313093d1d5 (MD5) Previous issue date: 2012-06-13 / Financiadora de Estudos e Projetos / The physical inactivity has been indicated by the World Health Organization (WHO) as one of the main risk factors for the incidence of Chronic Non-Communicable Diseases (CNCDs). Millions of deaths in the world are a result of these diseases, and this number has increased each year. In an attempt to change this scenario WHO has stimulated regular practice of physical activities, because they play an important role in preventing CNCDs. In Brazil, these activities are performed by health units which generate a large amount of data that need treatment. To deal with this problem we developed UCEMEPA, an environment that employs Ubiquitous Computing technologies and wireless communication networks, in order to monitor remotely and evaluate participants of physical activity groups in real-time. This environment automatically collects physiologic data, and provides indicators which will support and direct public policies for promoting physical activity. In this sense, UCEMEPA will contribute for the promotion of health and quality of life, and for the conduction of longitudinal studies aiming to establish correlations between the practice of physical activity and CNCDs prevention. / A inatividade física tem sido apontada pela Organização Mundial de Saúde (OMS) como um dos principais fatores de risco comportamentais responsáveis pela incidência de Doenças Crônicas Não Transmissíveis (DCNTs). Milhões de mortes no mundo são decorrentes dessas doenças e esse número vem aumentando a cada ano. Na tentativa de reverter esse quadro a OMS vem estimulando as práticas regulares de atividade física, já que estas possuem um importante papel na prevenção de DCNTs. No Brasil a promoção dessas atividades é realizada por unidades regionalizadas de saúde e geram uma grande quantidade de dados que carecem de processamento e tratamento. Em resposta a esse problema nós desenvolvemos o ACUMAAF, um ambiente que emprega tecnologias emergentes da Computação Ubíqua e redes de comunicação sem fio para monitorar e avaliar, em tempo real e a distância, participantes de grupos de atividade física. Esse ambiente coleta dados fisiológicos de forma automática e coletiva e tem como objetivo possibilitar a geração de indicadores capazes de apoiar e nortear políticas públicas de promoção de atividade física. O ACUMAAF é um ambiente computacional com contribuições para a promoção da saúde, para a promoção da qualidade de vida da população e para a realização de estudos longitudinais objetivando relacionar atividade física e a prevenção de DCNTs.