With the rapidly ageing world population, reform is needed to relieve the current health care system by a growing demand for interventions. The challenge is now to extend the period that elapses from the moment when a person, getting older, passes from independent living to the request of caregiving services. Assistive Technologies (AT) answer to this need with Smart Houses where Ambient Assisted Living Systems (AAL) is added to Home Automation (HA). A non-invasive monitoring system aims to prevent the occurrence of critical and difficult to diagnose situations such as dementia, and to improve the quality of independent living of elderly people and their families. We present in this thesis a method for detecting long-term behavioral changes (Behavioral Drift), in order to bring to the attention of specialists, data that can be used as support for early diagnosis. Among the first in this field, we will introduce an unsupervised method based on the evaluation of the Likelihood Ratio computed on a Hierarchical Markov Model in two distant time intervals. In order to get information about the activities of daily living (ADL) performed in the house, we have developed another unsupervised method. This Activity Recognition system is composed of a floating segmentation algorithm of the data relating to the sensors activation followed by k-means clustering. The design of methods and algorithms necessary to identify behavioral changes has required the use of large amounts of data. For this reason we have designed a system to convert and store data that exposes a standard interface for their usage. Finally, we created a new probabilistic model based on Mixture of Markov Renewal Processes to simulate the Home Automation response. This method, trained on real data (ARAS dataset), has been implemented and used in combination with SHARON, an ADL simulator, to generate the data that we used to validated the proposed system.
Con la popolazione mondiale in rapido invecchiamento, è necessaria una riforma per sgravare il corrente sistema sanitario da una crescente richiesta di interventi. La sfida attuale è quella di prolungare il periodo che intercorre dal momento in cui una persona, invecchiando, passa da una vita indipendente alla necessità di usufruire dell’assistenza dei caregivers. Le tecnologie assistive rispondo a questo bisogno con la fusione di Home Automation e Ambient Assisted Living in quelle che vengono oggi definite Case Intelligenti. Un sistema di monitoraggio non invasivo al fine di prevenire l’insorgenza di situazioni critiche e difficili da diagnosticare come la demenza, porterebbe ad un miglioramento della qualità della vita indipendente dell’anziano e della sua famiglia. Presentiamo in questa tesi un metodo per rilevare cambiamenti comportamentali (Behavioral Drift) che possono verificarsi in un lungo periodo, al fine di portare all’attenzione di specialisti, dati che possono essere di supporto per diagnosi precoci. Tra i primi in questo ambito, introdurremo l’utilizzo di un metodo non supervisionato basato sulla valutazione del Likelihood Ratio calcolato su un Modello Gerarchico di Markov in due distanti intervalli temporali. Al fine di ottenere informazioni circa le attività di vita quotidiana (ADL) svolte nella casa, abbiamo sviluppato un altro metodo, anch’esso non supervisionato. Tale sistema è composto da un algoritmo di segmentazione dei dati relativi all’attivazione di sensori seguito da K-means clustering. La sperimentazione di metodi e algoritmi atti ad identificare cambiamenti comportamentali ha richiesto l’utilizzo di una grande mole di dati. Per questo motivo abbiamo progettato un sistema per convertire e memorizzare dati che espone un’interfaccia standard per il loro utilizzo. Infine, abbiamo ideato un nuovo modello probabilistico per la simulazione della risposta domotica basato su Mixture of Renewal Markov Process e addestrato su dati reali (ARAS dataset). Tale metodo è stato implementato e utilizzato in combinazione con SHARON, un simulatore di ADL, per generare i dati con cui abbiamo validato il sistema proposto.
Unsupervised methods for activities of daily living drift modeling and recognition
TROFIMOVA, ANNA;MASCIADRI, ANDREA
2014/2015
Abstract
With the rapidly ageing world population, reform is needed to relieve the current health care system by a growing demand for interventions. The challenge is now to extend the period that elapses from the moment when a person, getting older, passes from independent living to the request of caregiving services. Assistive Technologies (AT) answer to this need with Smart Houses where Ambient Assisted Living Systems (AAL) is added to Home Automation (HA). A non-invasive monitoring system aims to prevent the occurrence of critical and difficult to diagnose situations such as dementia, and to improve the quality of independent living of elderly people and their families. We present in this thesis a method for detecting long-term behavioral changes (Behavioral Drift), in order to bring to the attention of specialists, data that can be used as support for early diagnosis. Among the first in this field, we will introduce an unsupervised method based on the evaluation of the Likelihood Ratio computed on a Hierarchical Markov Model in two distant time intervals. In order to get information about the activities of daily living (ADL) performed in the house, we have developed another unsupervised method. This Activity Recognition system is composed of a floating segmentation algorithm of the data relating to the sensors activation followed by k-means clustering. The design of methods and algorithms necessary to identify behavioral changes has required the use of large amounts of data. For this reason we have designed a system to convert and store data that exposes a standard interface for their usage. Finally, we created a new probabilistic model based on Mixture of Markov Renewal Processes to simulate the Home Automation response. This method, trained on real data (ARAS dataset), has been implemented and used in combination with SHARON, an ADL simulator, to generate the data that we used to validated the proposed system.File | Dimensione | Formato | |
---|---|---|---|
2015_12_Masciadri_Trofimova.pdf
accessibile in internet per tutti
Descrizione: Testo della tesi
Dimensione
7.68 MB
Formato
Adobe PDF
|
7.68 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/116330