Tesis sobre el tema "Detection and recognition of activities of daily living"

With the growing phenomenon of an aging population, an increasing numberof older people are living alone for domestic and social reasons. Based on thisfact, falling accidents become one of the most important factors in threateningthe lives of the elderly. Therefore, it is necessary to set up an application to de-tect the daily activities of the elderly. However, falling detection is difficult to recognize because the "falling" motion is an instantaneous motion and easy to confuse with others.In this thesis, three data mining methods were employed on wearable sensors' value; first which contains the continuous data set concerning eleven activities of daily living, and then an analysis of the different results was performed. Not only could the fall be detected, but other activities could also be classified. In detail, three methods including Back Propagation Neural Network, Support Vector Machine and Hidden Markov Model are applied separately to train the data set.What highlights the project is that a new  idea is put forward, the aim of which is to design a methodology of accurate classification in the time-series data set. The proposed approach, which includes obtaining of classifier parts and the application parts allows the generalization of classification. The preliminary results indicate that the new method achieves the high accuracy of classification,and significantly performs better than other data mining methods in this experiment.

La plupart des travaux réalisés dans le domaine de l'assistance à l'autonomie à domicile (AAL) reposent sur l'utilisation de capteurs visuels et audio tels que des caméras. Or, ces capteurs sont souvent rejetés par le patient à cause de leur caractère invasif. Des approches alternatives requièrent l'utilisation de capteurs embarqués sur la personne (GPS, bracelets électroniques ou puces RFID dans les vêtements...), et leur pertinence est donc ramenée à l'hypothèse que les personnes les portent effectivement, sans jamais les rejeter ni les oublier. Pour ces raisons, dans cette thèse, nous trouvons plus pertinentes les approches uniquement basées sur l'utilisation de capteurs binaires intégrés dans l'habitat, tels que les détecteurs de mouvements, les tapis sensitifs ou les barrières optiques. Dans un tel contexte technologique, il devient intéressant d'utiliser les paradigmes, les modèles et les outils des systèmes à événement Discrets (SED), initialement plutôt développés pour la modélisation, l'analyse et la commande des systèmes industriels complexes. Dans ces travaux de thèse, l'objectif est de construire une approche pour la modélisation et le suivi des habitudes de vie, basée sur les modèles et les paradigmes des SED et répondant à une problématique qui s'énonce de la manière suivante : L'objectif est de développer un cadre global pour découvrir et reconnaître les activités de la vie quotidienne d'un habitant vivant seul dans une maison intelligente. Cette maison intelligente doit être équipée uniquement de capteurs binaires, l'étiquetage par des experts des activités observées ne doit pas être nécessaire et les activités peuvent être représentées par des modèles probabilistes. La première méthode présentée dans cette thèse permet, à partir d'une base de données d'apprentissage et d'une description experte des activités à modéliser listées par des médecins, de construire pour chaque activité un modèle sous la forme d'un automate à état-fini probabiliste (PFA). La seconde méthode développée lors de cette thèse permet d'estimer en temps réel, à partir des seules données observées par les capteurs ambiants, quelle activité la personne observée réalise effectivement. Les méthodes décrites dans cette thèse sont illustrées en utilisant les données générées localement via l'utilisation d'un appartement prêté par ENS Paris-Saclay équipé pour répondre aux besoins expérimentaux de cette thèse
Most of the work done in the field of ambient assisted living (AAL) is based on the use of visual and audio sensors such as cameras. However, these sensors are often rejected by the patient because of their invasiveness. Alternative approaches require the use of sensors embedded on the person (GPS, electronic wristbands or RFID chips in clothing ...), and their relevance is therefore reduced to the assumption that people actually wear them, without rejecting nor forgetting them. For these reasons, in this thesis, we find more relevant the approaches based on the use of binary sensors integrated into the habitat only, such as motion detectors, sensory mats or optical barriers. In such a technological context, it becomes interesting to use paradigms, models and tools of Discrete Event Systems (DES), initially developed for modeling, analysis and control of complex industrial systems. In this thesis work, the goal is to build an activity of daily living modeling and monitoring approach, based on the models and the paradigm of the DES and answering a problem that is expressed as follows:The objective is to develop a global framework to discover and recognise activities of daily living of an inhabitant living alone in a smart home. This smart home have to be equipped with binary sensors only, expert labeling of activities should not be needed and activities can be represented by probabilistic models. The first method presented in this thesis allows to build a probabilistic finite-state automata (PFA) from a learning database and an expert description of the activities to be modeled given by the medical staff. The second method developed during this thesis estimates, according to the observations, the activity performed by the monitored inhabitant. The methods described in this thesis are applied on data generated using an apartment lent by ENS Paris-Saclay and equipped according the experimental needs of this thesis
La maggior parte dei lavori nel settore dell’Ambient Assisted Living (AAL) si basasull’uso di sensori visivi e audio come le telecamere. Tuttavia, questi sensori sonospesso rifiutati dal paziente a causa della loro natura invasiva. Gli approcci alternativi richiedono l’uso di sensori integrati nella persona (GPS, bracciali elettronici o chipRFID...), e la loro rilevanza è quindi ridotta all’ipotesi che le persone li indossino effettivamente, senza mai rifiutarli o dimenticarli.Per questi motivi, in questa tesi, troviamo approcci più rilevanti basati esclusivamente sull’uso di sensori binari integrati nell’habitat, come rilevatori di movimento,tappeti sensoriali o barriere fotoelettriche.In tale contesto tecnologico, diventa interessante utilizzare i paradigmi, i modelli egli strumenti dei sistemi ad eventi discreti (SED), inizialmente sviluppati per la modellazione, l’analisi e il controllo di sistemi industriali complessi.In questo lavoro di tesi, l’obiettivo è quello di presentare un metodo per la modellazione e il monitoraggio delle abitudini di vita, basato sui modelli e paradigmi di SEDe rispondendo ad un problema che si esprime come segue : L’obiettivo è quello di sviluppare un quadro globale per rivelare e riconoscere le attività della vita quotidiana di una persona che abita da sola in una smart home chedovrebbe essere dotata solo di sensori binari. Inoltre si suppone che non sia necessarial’etichettatura delle attività osservate da parte di un esperto e tali attività sono rappresentate da modelli probabilistici.Il primo metodo presentato in questa tesi permette di costruire un modello probabilistico di automa a stati finiti (PFA) ottenuto da un database di apprendimento e unadescrizione delle attività da parte di medici. Il secondo metodo sviluppato in questa tesi stima, alla luce delle osservazioni, qualeattività svolge la persona osservata. I metodi descritti sono illustrati utilizzando dati generati localmente attraverso l’usodi un appartamento messo a disposizione da ENS Paris-Saclay e attrezzato per soddisfarele esigenze sperimentali di questa tesi

Due to the rise in the elderly population and the prevalence of chronic diseases, healthcare organizations around the world are faced with an economic burden which will continue to grow. For this reason there is an urgent demand to reduce the intake of elders in hospitals and nursing homes by allowing them to live independently for greater lengths of time. In response to this demand, researchers are strongly focusing on 'telemonitoring', which is the use of information technology (IT) to monitor the health status of an individual from a remote location (e.g. their home). The first aim of this thesis was to investigate how telemonitoring technologies can detect elderly activities for health assessment purposes. To assess a patient's health status holistically, a wide variety of factors needed be considered by practitioners. Currently many telemonitoring technologies in research have addressed these factors/assessments by monitoring elderly activities. However based on the literature reviews, researchers have not been able to develop a comprehensive understanding of how these technologies support each assessment. Therefore the first contribution in chapter 3 of this thesis addresses this gap. A literature review was conducted where 215 telemonitoring technologies were identified from 82 papers, published between 2000 and 2016. Six assessments which involve monitoring of activities were identified as (1) mobility, (2) nutrition, (3) safety, (4) cognitive, (5) social, and (6) routine. All included technologies were categorized into six tables according to the assessment that they supported. From assessing the contents of these tables, it was found that a significant portion of ITs relate to mobility, nutritional, safety and routine assessments. Many of the studies were found to assess technologies inside of laboratory setting and still require improvement before they are suited for real world application. It also found that many of the technologies were not equipped with wireless communication. In recent years recently have been focusing been integrating wireless sensing technology into telemonitoring applications. Many of these wireless technologies are small, unobtrusive, and usually need to be powered by small batteries (e.g. coin cell) which have limited capacity. For this reason, researchers have had difficulties prolonging battery life to a duration that is practical. However the recent release of Bluetooth Low Energy (BLE) has the potential of resolving this issue due to its power saving qualities. The 2nd aim of this thesis is to assess the performance of Bluetooth Low Energy (BLE) in telemonitoring frameworks using advertising mode. Advertising mode is often used for device discovery purposes, however it can also be used to send context data without the need for device connection establishment. This method has received little investigation from researchers and yet it has the potential of offering advantages such as reduction in power consumption and manufacturing costs. Therefore in this thesis, the performance of BLE advertising mode was used within two telemonitoring applications. Firstly, a new device called 'BLUESOUND' is proposed. The device uses ultrasound sensing technology which can efficiently differentiate multiple residents in a home environment based on their height. The device consists of three sensing/communication modules: A Passive Infrared (PIR) occupancy module, an ultrasound array module and a BLE communication module. The PIR occupancy module is used to detect walking direction, while the ultrasound array measures the resident's height. The combination of these two technologies can also be used to detect a resident's velocity. BLE advertising mode is used to communicate acquired data to a smart phone gateway/database. A new embedded algorithm was able to increase the energy efficiency of the identification technology. Comprehensive modelling and experimentation was undertaken to assess the performance the BLUESOUND device. The BLUESOUND device was able to distinguish between multiple resident identities by measuring height accurately. Currently researchers have developed various wearable ECG monitors as there is a demand to detect the onset of cardiac disease earlier in the elderly population. However most of these devices have only lasted a couple of days on a coin cell battery which is not practical. Therefore the performance of BLE advertising mode was explored using a virtual BLE based ECG model in MATLAB. To further minimize power consumption, an ECG extraction technique (based on the 'So and Chan' algorithm [1]) was used in the model to extract some of the most significant points on the signal. Based on three simulation trails, ECG data was transferred to a scanning device with high accuracy (average of 99.62%). It was estimated that the virtual system is approximately 13 times more energy efficient compared to sending ECG stream data continuously when a connection is established.

Activity recognition is becoming an increasingly important task in artificial intelligence. Successful activity recognition systems must be able to model and recognise activities ranging from simple short activities spanning a few seconds to complex longer activities spanning minutes or hours. We define activities as a set of qualitatively interesting interactions between people, objects and the environment. Accurate activity recognition is a desirable task in many scenarios such as surveillance, smart environments, robotic vision etc. In the domain of robotic vision specifically, there is now an increasing interest in autonomous robots that are able to operate without human intervention for long periods of time. The goal of this research is to build activity recognition approaches for such systems that are able to model and recognise simple short activities as well as complex longer activities arising from long-term autonomous operation of intelligent systems. The research makes the following key contributions: 1. We present a qualitative and quantitative representation to model simple activities as observed by autonomous systems. 2. We present a hierarchical framework to efficiently model complex activities that comprise of many sub-activities at varying levels of granularity. Simple activities are modelled using a discriminative model where a combined feature space, consisting of qualitative and quantitative spatio-temporal features, is generated in order to encode various aspects of the activity. Qualitative features are computed using qualitative spatio-temporal relations between human subjects and objects in order to abstractly represent the simple activity. Unlike current state-of-the-art approaches, our approach uses significantly fewer assumptions and does not require any knowledge about object types, their affordances, or the constituent activities of an activity. The optimal and most discriminating features are then extracted, using an entropy-based feature selection process, to best represent the training data. A novel approach for building models of complex long-term activities is presented as well. The proposed approach builds a hierarchical activity model from mark-up of activities acquired from multiple annotators in a video corpus. Multiple human annotators identify activities at different levels of conceptual granularity. Our method automatically infers a ‘part-of’ hierarchical activity model from this data using semantic similarity of textual annotations and temporal consistency. We then consolidate hierarchical structures learned from different training videos into a generalised hierarchical model represented as an extended grammar describing the over all activity. We then describe an inference mechanism to interpret new instances of activities. Simple short activity classes are first recognised using our previously learned generalised model. Given a test video, simple activities are detected as a stream of temporally complex low-level actions. We then use the learned extended grammar to infer the higher-level activities as a hierarchy over the low-level action input stream. We make use of three publicly available datasets to validate our two approaches of modelling simple to complex activities. These datasets have been annotated by multiple annotators through crowd-sourcing and in-house annotations. They consist of daily activity videos such as ‘cleaning microwave’, ‘having lunch in a restaurant’, ‘working in an office’ etc. The activities in these datasets have all been marked up at multiple levels of abstraction by multiple annotators, however no information on the ‘part-of’ relationship between activities is provided. The complexity of the videos and their annotations allows us to demonstrate the effectiveness of the proposed methods.

There is an increasing concern in tackling the problems faced by the elderly community and physically in-locked people to lead an independent life experience problems with self- care. The need for developing service robots that can help people with mobility impairments is hence very essential. Developing a control framework for shared human-robot autonomy will allow locked-in individuals to perform the Activities of Daily Living (ADL) in a exible way. The relevant ADL scenarios were identi ed as handling objects, self-feeding, and opening doors for indoor nav- igation assistance. Multiple experiments were conducted, which demonstrates that the robot executes these daily living tasks reliably without requiring adjustment to the environment. The indoor manipulation tasks hold the challenge of dealing with a wide range of unknown objects. This thesis presents a framework developed for grasping without requiring a priori knowledge of the objects being manipulated. A successful manipulation task requires the combination of aspects such as envi- ronment modeling, object detection with pose estimation, grasp planning, motion planning followed by an e?cient grasp execution, which is validated by a 6+2 Degree of Freedom robotic manipulator.

This thesis studies novel outdoor and indoor positioning technologies enhanced with wearable or body worn motion sensors to recognize daily activities. In gerontology and geriatric medicine, it is fundamentally important to be able to detect changes in physical, psychosocial, and cognitive outcomes over time. In this thesis, intelligent systems for assessing aging changes were discussed. By utilizing novel technologies and methods, efficiency of home care services can be improved. The studied concept enables early detection of changes in functional ability and daily performance. As a research methods both interviews and literature studies were used. The technology study was conducted as a literature study, whereas the concept creation and selection of suitable technologies were based on interviews with experts belonging to 5GTN alliance. As a result, technologies for outdoor and indoor tracking were selected, and sensors for real time tracking of the daily activities and routines were proposed. Pilot project of indoor tracking of elderly people is starting in autumn 2017. Results of that are not covered in this thesis but those are covered in forthcoming theses
Opinnäytetyössä tutkittiin ulko- ja sisäpaikannusteknologioita, sekä yhdessä niiden kanssa käytettäviä kannettavia ja/tai puettavia liikesensoreita, joilla voidaan havaita ja tunnistaa päivittäisiä aktiviteettejä. Ikääntyvien ihmisten hoidossa on erinomaisen tärkeää pystyä havaitsemaan muutoksia fyysisessä, psykososiaalisessa sekä kognitiivisessa toimintakyvyssä. Tässä työssä käsiteltiin älykkäitä järjestelmiä ikääntymisestä aiheutuneiden muutosten arvioimiseksi. Uusia teknologioita ja menetelmiä hyödyntämällä voidaan parantaa kotihoitopalvelujen tehokkuutta. Työssä tutkittu konsepti mahdollistaa toimintakyvyn muutosten, sekä päivittäisen suorituskyvyn muutosten varhaisen tunnistamisen. Tutkimusmenetelminä käytettiin sekä haastatteluja että kirjallisuustutkimuksia. Teknologiatutkimus suoritettiin kirjallisuustutkimuksena eri lähteistä, kun taas konseptointi ja teknologioiden valinta suoritettiin haastattelemalla 5GTN allianssiin kuuluvia moniammatillisia asiantuntijoita. Tutkimuksen tuloksena valittiin käytettävät teknologiat ulko- ja sisäpaikannusmenetelmiin, sekä valittiin sensorityypit päivittäisten toimintojen ja rutiinien reaaliaikaiseen seurantaan. Syksyllä 2017 on käynnistymässä pilottiprojekti, jossa sisätila-antureiden avulla seurataan ikäihmisten toimintaa omassa kodissaan. Tuloksia tuosta projektista ei käsitellä tässä opinnäytetyössä, vaan ne käsitellään tulevissa opinnäytteissä

La reconnaissance d'activités de vie quotidienne (AVQs) fait l'objet, depuis des années, de recherches pour proposer des solutions performantes. Elle s'appuie sur l'analyse spatio-temporelle de situations, de comportements, etc. dont les données d'entrée sont les informations fournies par des capteurs ambiants ou par des capteurs portés par la personne. Cette thèse se focalise sur l'instrumentation du lieu de vie par des capteurs ambiants et sur la détection d'un état d'isolement social chez les personnes âgées. Deux approches sont utilisées pour évaluer l'isolement social. La première se base sur des questionnaires. La deuxième approche s'appuie sur l'utilisation de technologies pour la récupération objective de données représentatives d'un état ou d'un comportement. En particulier, l'activité " se nourrir " est liée à une séquence d'actions telles que faire les courses, cuisiner, manger et faire la vaisselle. L'activité " se déplacer " est liée à la mobilité au sein du domicile et au fait de sortir du logement. La littérature montre que ces deux activités semblent pertinentes pour évaluer un risque potentiel d'isolement social chez les aînés. Les travaux de thèse portent sur quatre contributions principales : Un état bibliographique des recherches sur la détection des AVQs afin d'en identifier les apports et les limites et tracer des voies de recherches pertinentes. Des critères spécifiques ont été choisis pour inclure les articles dans lesquels des systèmes de détection d'activités sont présentés. Une démarche de conception système appliquée à la reconnaissance d'AVQs. Cette démarche s'intègre dans un processus d'Ingénierie Système. Elle décrit l'analyse des exigences, leur modélisation au travers de diagrammes SysML et la mise en place d'une architecture matérielle et logicielle basée sur un réseau IoT. L'analyse des AVQs, dans notre étude, utilise les données de détecteurs de mouvement et de capteurs de contacts. Un affichage sur une application web permet de visualiser les résultats obtenus à destination de l'aide-soignant ou de la famille. L'utilisation originale de quatre méthodes de classification des AVQs à savoir "préparer le repas", "prendre le repas", "faire la vaisselle", "dormir/se relaxer", "hygiène", "la personne à l'extérieur du logement", "un visiteur à l'intérieur de la maison" et "autres activités". Les trois premières méthodes utilisées sont K-means, le modèle de mélange gaussien et BIRCH auxquelles on applique une pondération aux données. Les activités liées au repas n'ont ainsi pas le même poids que le reste des données, ce qui a permis d'améliorer la détection des AVQs. Le quatrième algorithme est basé sur une méthode logique à la suite de la détermination d'une matrice de corrélation prenant en entrée l'ensemble des capteurs disponibles. En utilisant les données de la matrice de corrélation, l'algorithme personnalise la détection des activités liées au repas en différenciant une personne qui prépare seule son repas d'une personne qui bénéficie d'un service de portage de repas. Nous validons nos méthodes en se référant aux formulaires remplis par les participants au début et à la fin de l'expérimentation dans lesquels ils indiquent le déroulement de leur journée-type. Ces algorithmes ont été appliqués sur une base de données annotée ouverte pour confirmer la précision de nos approches. La proposition d'un score du niveau d'isolement social chez la personne suivie. Ce score est établi sur la base de l'identification des activités pour extraire les habitudes quotidiennes au travers d'indicateurs du comportement (Le temps passé à l'extérieur de la maison et à l'intérieur de la cuisine, etc.). Six aînées ont été suivies pendant plus de 3 mois. L'algorithme régression logistique a été utilisé pour l'extraction du niveau d'isolement social qui a été comparé à celui identifié grâce au questionnaire " Lubben Social Network Scale " rempli avec chaque participant au début et à la fin de l'étude
The recognition of daily life activities has been the subject of research for years to provide effective solutions. It is based on the spatio-temporal analysis of situations and behaviors whose input data is information provided by ambient sensors or by sensors worn by the person. This thesis focuses on the instrumentation of the living space by ambient sensors and on the detection of a state of social isolation in elderly people. Two approaches are used to assess social isolation. The first one is based on questionnaires. The second approach is based on the use of technologies for the objective acquisition of data representative of a state, behavior, etc. In particular, the activity "eating" is linked to a sequence of actions such as shopping, cooking, eating and washing dishes. The activity "moving" is linked to mobility within the home and leaving the home. The literature shows that these two activities seem to be relevant for assessing a potential risk of social isolation among older people. The thesis work focuses on four main contributions: A bibliographic review of ADLs detection research to identify its contributions and limitations, and to outline relevant research directions. Specific criteria were chosen to include articles presenting activity detection systems. A system design approach applied to the detection of ADLs. This approach is part of a system engineering process. It describes the analysis of requirements, their modeling through SysML diagrams and the implementation of a hardware and software architecture based on an IoT network. The analysis of ADLs, in our study, uses data from motion detectors and contact sensors. A display on a web application allows you to visualize the results obtained for the caregiver or the family. The original use of four methods to classify ADLs, namely "preparing the meal", "eating the meal", "washing the dishes", "sleeping/relaxing", "hygiene", "the person outside the home", "a visitor inside the home" and "other activities". The first three methods used are K-means, the Gaussian mixture model and BIRCH, which applies weights to the data. Meal-related activities therefore do not have the same weight as the rest of the data, which made it possible to improve the detection of ADLs. The fourth algorithm is based on a logical method following the determination of a correlation matrix using all the available sensors as input. Using the data from the correlation matrix, the algorithm personalizes the detection of meal-related activities by distinguishing a person preparing their meal from a person using a meal delivery service. We validate our methods by referring to the forms filled in by the participants at the beginning and end of the experiment, in which they describe the course of their typical day. These algorithms were applied to an open annotated database to confirm the accuracy of our approaches. The proposal of a score for the level of social isolation of the person being monitored. This score is based on the identification of activities to extract daily habits through behavioral indicators (time spent outside the house and in the kitchen, etc.). Six elderly people were followed for more than 3 months. The logistic regression algorithm was used to extract the level of social isolation, which was compared with the level of social isolation identified using the "Lubben Social Network Scale" questionnaire, which was completed by each participant at the beginning and end of the study

Nowadays, with the rapid development of the Internet of Things, the applicationfield of wearable sensors has been continuously expanded and extended, especiallyin the areas of remote electronic medical treatment, smart homes ect. Human dailyactivities recognition based on the sensing data is one of the challenges. With avariety of data mining techniques, the activities can be automatically recognized. Butdue to the diversity and the complexity of the sensor data, not every kind of datamining technique can performed very easily, until after a systematic analysis andimprovement. In this thesis, several data mining techniques were involved in theanalysis of a continuous sensing dataset in order to achieve the objective of humandaily activities recognition. This work studied several data mining techniques andfocuses on three of them; Decision Tree, Naive Bayes and neural network, analyzedand compared these techniques according to the classification results. The paper alsoproposed some improvements to the data mining techniques according to thespecific dataset. The comparison of the three classification results showed that eachclassifier has its own limitations and advantages. The proposed idea of combing theDecision Tree model with the neural network model significantly increased theclassification accuracy in this experiment.

The need for greater independence amongst the growing population of elderly people has made the concept of “ageing in place” an important area of research. Remote home monitoring strategies help the elderly deal with challenges involved in ageing in place and performing the activities of daily living (ADLs) independently. These monitoring approaches typically involve the use of several sensors, attached to the environment or person, in order to acquire data about the ADLs of the occupant being monitored. Some key drawbacks associated with many of the ADL monitoring approaches proposed for the elderly living alone need to be addressed. These include the need to label a training dataset of activities, use wearable devices or equip the house with many sensors. These approaches are also unable to concurrently monitor physical ADLs to detect emergency situations, such as falls, and instrumental ADLs to detect deviations from the daily routine. These are all indicative of deteriorating health in the elderly. To address these drawbacks, this research aimed to investigate the feasibility of unsupervised monitoring of both physical and instrumental ADLs of elderly people living alone via inexpensive minimally intrusive sensors. A hybrid framework was presented which combined two approaches for monitoring an elderly occupant’s physical and instrumental ADLs. Both approaches were trained based on unlabelled sensor data from the occupant’s normal behaviours. The data related to physical ADLs were captured from Kinect sensors and those related to instrumental ADLs were obtained using a combination of Kinect sensors and a power meter. Kinect sensors were employed in functional areas of the monitored environment to capture the occupant’s locations and 3D structures of their physical activities. The power meter measured the power consumption of home electrical appliances (HEAs) from the electricity panel. A novel unsupervised fuzzy approach was presented to monitor physical ADLs based on depth maps obtained from Kinect sensors. Epochs of activities associated with each monitored location were automatically identified, and the occupant’s behaviour patterns during each epoch were represented through the combinations of fuzzy attributes. A novel membership function generation technique was presented to elicit membership functions for attributes by analysing the data distribution of attributes while excluding noise and outliers in the data. The occupant’s behaviour patterns during each epoch of activity were then classified into frequent and infrequent categories using a data mining technique. Fuzzy rules were learned to model frequent behaviour patterns. An alarm was raised when the occupant’s behaviour in new data was recognised as frequent with a longer than usual duration or infrequent with a duration exceeding a data-driven value. Another novel unsupervised fuzzy approach to monitor instrumental ADLs took unlabelled training data from Kinect sensors and a power meter to model the key features of instrumental ADLs. Instrumental ADLs in the training dataset were identified based on associating the occupant’s locations with specific power signatures on the power line. A set of fuzzy rules was then developed to model the frequency and regularity of the instrumental activities tailored to the occupant. This set was subsequently used to monitor new data and to generate reports on deviations from normal behaviour patterns. As a proof of concept, the proposed monitoring approaches were evaluated using a dataset collected from a real-life setting. An evaluation of the results verified the high accuracy of the proposed technique to identify the epochs of activities over alternative techniques. The approach adopted for monitoring physical ADLs was found to improve elderly monitoring. It generated fuzzy rules that could represent the person’s physical ADLs and exclude noise and outliers in the data more efficiently than alternative approaches. The performance of different membership function generation techniques was compared. The fuzzy rule set obtained from the output of the proposed technique could accurately classify more scenarios of normal and abnormal behaviours. The approach for monitoring instrumental ADLs was also found to reliably distinguish power signatures generated automatically by self-regulated devices from those generated as a result of an elderly person’s instrumental ADLs. The evaluations also showed the effectiveness of the approach in correctly identifying elderly people’s interactions with specific HEAs and tracking simulated upward and downward deviations from normal behaviours. The fuzzy inference system in this approach was found to be robust in regards to errors when identifying instrumental ADLs as it could effectively classify normal and abnormal behaviour patterns despite errors in the list of the used HEAs.

Les progrès de la technologie des capteurs et leur disponibilité ont permis de mesurer diverses propriétés et activités des habitants dans une maison intelligente. Cependant, l’obtention de connaissances significatives à partir d’une grande quantité d’informations collectées à partir d’un réseau de capteurs n’est pas une tâche simple. En raison de la complexité du comportement des habitants, l’extraction d’informations significatives et la prédiction précise des valeurs représentant les activités futures d’un occupant sont des défis de recherche [6]. L’objectif principal de notre travail de thèse est d’assurer une analyse efficace des données recueillies à partir des capteurs d’occupation dans une maison intelligente. A ce propos, ce travail se base sur la reconnaissance et l’évaluation des activités quotidiennes d’une personne âgée dans le but d’observer, de prédire et de suivre l’évolution de son état de dépendance, de santé et de détecter par la même occasion, la présence d’une perte ou d’une perturbation de l’autonomie en temps réel
Advances in sensor technology and their availability have measured various properties and activities of residents in a smart home. However, obtaining significant knowledge from a large amount of information collected from a sensor network is not a simple task. Due to the complexity of the behavior of the inhabitants, the extraction of meaningful information and the accurate prediction of values ​​representing the future activities of an occupant are research challenges. The main objective of our thesis work is to ensure an efficient analysis of data collected from occupancy sensors in a smart home. In this regard, this work is based on the recognition and evaluation of the daily activities of an elderly person in order to observe, predict and monitor the evolution of his state of dependence, health and to detect by the same occasion, the presence of a loss or a disruption of autonomy in real time

L'augmentation de la population séniore se révèle être un enjeu de taille et une grande question de santé publique. La part démographique des personnes âgées s'agrandit de plus en plus grâce au progrès et avancés de la médecine et nos systèmes de santé. Néanmoins le vieillissement de cette population implique naturellement une pléthore de problèmes de dépendance qui y sont associés, ceci bien sûr exponentiellement. Les maisons de retraite sont des solutions généralement coûteuses et très peu appréciées. En conséquence, des solutions plus adaptées basées sur l'aide au maintien à domicile ce développement de plus en plus ces dernières années. Cette problématique se retrouve dans la croisée des chemins entre les technologies de capteurs, la télétransmission de données, l'assistance aux personnes âgées à mobilité réduite et le suivi d'activités. Cette thèse explore l'application d'algorithmes d'analyse de données pour le suivi d'activités des personnes âgées à domicile. L'idée étant qu'un suivi régulier des résidents permet d'inférer leur état de dépendance ou d'autonomie et permet aux personnels soignants d'intervenir en cas de détection d'un début de dégradation. Nous avons exploré et adapté certaines méthodes d'inférence bayésienne et segmentation de séries temporelles pour la reconnaissance d'activités. Et ensuite, nous avons proposé un outil de visualisation permettant de faciliter la détection d'anomalies ou changements de rythme de vie. Ce travail s'inscrit dans le cadre d'une thèse CIFRE. Ainsi tous les méthodes et algorithmes explorés ont été mis en production et sont exploités par la solution d'aide à domicile commercialisé par la société Diatelic
The increase of the senior population constitutes a major public health issue. The demographic share of the elderly is ever more growing thanks to the progress and advances in medicine and our health care systems. However, with the aging of this population comes a plethora of dependency problems, and this, of course, exponentially.Retirement homes are an expensive and not very popular solution. As a result, we are seeing a surge in home assisted living solutions in the recent years.This topic is in the crossroads between sensor technologies, data transmission, assistance to elderly people and activity monitoring.This thesis explores the application of data analysis algorithms for activity monitoring of elderly people at home. The idea is that with day-to-day monitoring of residents it is possible to infer their autonomy and capacity to perform day-to-day tasks. It also allows caregivers to intervene in cases where the start of some degradation is detected.We explored and adapted some Bayesian inference and time series segmentation methods for activity recognition. And then, we proposed a visualization tool to facilitate the detection of anomalies or changes in everyday habits.This work is part of a CIFRE thesis. The methods and algorithms presented have been put into production and are packaged into Diatelic's the assisted living commercial solution

碩士
國立臺灣大學
資訊工程學研究所
98
We combine the camera, the sound recorder, and the accelerometer on the smart-phone with the machanism of SenseCam to invent a new application of smartphones called as SenseMobile. SenseMobile employs images, sounds and accelerometer values to build an activities of daily living(ADLs) recognition system. However other portable devices merely recognize physical activities instead of high-level activities. In this thesis, we extract effective features to implementing activity recognition. In image feature extraction, we detect human face and cluster local images after pre-processing. For sound feature extraction, in the time domain, we extract volume, non-silent ratio and two human voice features - maximum peak value and number of peaks. Furthermore, From the frequency domain, we extract Mel-frequency cepstral coefficients (MFCCs), which are popular in speech recognition. After clustering vibration types, we calculate probabilities of types in accelerometer feature extraction. Then we sample instances on sliding time window and implement classification on machine learning models. We design two experiments - ADLs recognition in experimental environment and in real environment. In multiple classifications, we compare accuracy from Support Vector Machine(SVM) and Hidden Markov Model(HMM) models, and from distinct data types. In binary classifications, we utilize one-against-all method and optimize individual activity recognition. Eventually, results of two experiments prove success in ADLs recognition and bring forward unsolved defects of SenseMobile.

The healthcare systems are experiencing heavy workload and high cost caused by ageing population. The assisted monitoring systems for the elderly persons, and persons with chronic diseases, promises great potential to provide them with care and comfort at the privacy of their own homes and as a result help reduce healthcare costs. This requires a monitoring system capable of detecting daily human activities in living spaces. In this work we discuss different challenges to design such a system, present an activity data visualization tool designed to study human activities in a living space and propose a two stage, supervised statistical model for detecting the activities of daily living (ADL) from non-visual sensor data streams. A novel data segmentation is proposed for accurate prediction at the first stage. We present a novel error correction structure for the second stage to boost the accuracy by correcting the misclassification from the first stage.

碩士
國立交通大學
電控工程研究所
100
Human activity recognition system is now a very popular subject for research and application. Using a fixed camera to track a person and recognize his (her) activity is widely seen in home surveillance. For real-time surveillance, the embedded algorithms must be efficient and fast to meet the real-time constraint. In the thesis, a new person tracking and continuous activity recognition is proposed. We build two background models, in grayscale and HSV color space as well to extract the human correctly, and we could also reduce the shadowing effect well. For better efficiency and separability, the binary image is firstly transformed to a new space by eigenspace and then canonical space transformation, and the recognition is finally done in canonical space. A three image frame sequence, 5:1 down sampling from the video, is converted to a posture sequence by template matching. The posture sequence is classified to an action by fuzzy rules inference. Fuzzy rule approach can not only combine temporal sequence information for recognition but also be tolerant to variation of action done by different people and time. Moreover, we make use of the hue component to recognize the medical pouch’s color when one is taking medicine. By combining with the hue-based pouch’s color model and human activity recognition system, we can know someone is taking medicine and its medical pouch’s color as well. Finally, we also employ the activity recognition system to record a student’s activity in the daily living.

碩士
國立中央大學
資訊工程學系
105
In recent years, because of the improvement of medical technology, Taiwan is facing the severe problem of population aging. Since young people move out for work or marriage, the health care of independent-living Elderly is more important than ever. How to measure the activities of daily living for the elderly in an effective way is the crucial issue nowadays. In this paper, we developed a DNN-based System for the Recognition of the Activities of Daily Living. The system estimates skeleton data from the color image, which is recorded from webcam or surveillance system, and using the neural network like CNN, BPN or DNN to classify these features proposed by this paper. After recognized motions, we log the data in order to give the user a daily report. In this paper, we design ten different activities of daily living including one Scene of falling movement, and testing these data with angular tolerance and person independent experiments. In these experiments, we obtained a great result of over 90% recognition rate. Even in the real-life test, this system precision rate can also achieve 92.93%. With these experiments, we can prove that the system is good enough to provide a robust report to the user for consulting.

The dissertation thesis focuses on the topic of construction of new questionnaire, which is developed from four already existing questionnaires of activities of daily living. This new questionnaire summarizes strengths of those four and submits only those activities, which are affected at the early stage of Alzheimer's dementia. The topic is introduced in the broad context of social sciences, medical science and also submits the practical issues, considering effects of Alzheimer's disease to life of patient and his closest ones. New questionnaire, which is the result of empirical research, is based on correlation analysis of three commonly used screening cognitive tests and four questionnaires which evaluate activities of daily life, which are, used as best practice in the Counseling Center for Memory Disorders Clinic of Neurology of Faculty Hospital Kralovske Vinohrady in Prague. The research sample consists of patiens, who are affected with Alzheimer's disease already. That means, that the new questionnaire can be used as a tool for those, who are the patient's closest ones for purpose of screening and reviewing abilities of daily living over time. The questionnaire also presents comprehensively compiled a list of activities of daily living which Alzheimer's disease affects in early stage and this list...