| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.44 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Com o aumento da esperança média de vida, é percetível o envelhecimento da população e, consequentemente, uma diminuição das condições sociais e económicas para o cuidado diário de idosos. Tendo isto em conta, mais atenção está a ser dada aos sistemas de cuidados remotos para ajudar os pacientes a cuidar de si mesmos, diminuindo a necessidade de cuidados de saúde convencional. A atividade física humana é uma característica importante de alguns dos novos sistemas de prevenção e monitorização, portanto parâmetros relacionados com esta temática têm recebido um interesse crescente por parte dos profissionais de saúde, como, por exemplo, fisioterapeutas, cuidadores de idosos e nutricionistas.
Este projeto propõe-se a detetar vários modos de transporte, usando para isso informação proveniente dos vários sensores disponíveis num smartphone comum, adicionando ainda métodos que se aproveitam das informações circundantes, tais como pontos de acesso Wi-Fi de forma a melhorar a precisão global do sistema.
Ao monitorizar os modos mais comuns de transporte utilizados pela população sénior, como caminhar, autocarro, metro, comboio e carro, será possível auxiliar o trabalho dos profissionais de saúde, dando-lhes conhecimento sobre os hábitos de atividade física do utilizador, permitindo ainda perceber se o idoso tem uma vida sedentária, se existem mudanças nos padrões de movimento diário, fornecer indicação sobre a falta de atividades sociais, se é suficientemente independente na sua vida quotidiana, ou se está a fazer exercício suficiente.
Esta tese descreve o modo como vários classificadores referidos no estado da arte para a deteção do modo de transporte foram treinados e avaliados. O classificador, Decision Tree, demonstrou melhor performance na deteção do modo de transporte e foi implementado de forma a poder ser utilizado por outras aplicações Android. Para isso foi implementado um sistema como prova de conceito para demonstrar o funcionamento do classificador de transporte. O classificador foi incorporado numa aplicação para o cuidado de idosos, que pode ser usado por profissionais de saúde e afins para monitorizar os padrões de atividade diária das pessoas idosas.
O classificador foi criado usando mais de 24 horas de dados de transporte de um grupo de 15 indivíduos e pode, teoricamente, atingir mais de 96,1% de precisão. No entanto, uma validação do mundo real do sistema implementado obteve 88,97% de precisão. Em alinhamento com a investigação feita nesta tese, um artigo foi submetido e aceite para conferência. O anexo C apresenta o trabalho aceite na octogésima Conferência Internacional IEEE Healthcom, realizada em Munique, em setembro de 2016.
With an increase in average life expectancy, the aging of the population is discernible and consequently a reduction in social and economic conditions for elderly daily care. Taking this into account, more attention is being given to remote care systems to assist the patients and help them take care of themselves, lowering the conventional health care necessity. Human physical activity monitoring is an important feature of such systems, and has received an increasing interest from health-related professionals, such as physiotherapists, elders’ caregivers and nutritionists. This project’s main purposes are the detection of transport modes (walking, bus, car, metro, ...) using data from different sensors available on a common smartphone, and the addition of methods that take advantage of the surrounding environment, such as Wi-Fi Access Points, in order to improve the global accuracy of the system. By monitoring the most common modes of transport used by this population it is possible to support the work of caregivers, giving them knowledge about the user's activity habits, perceive whether the elder is having a sedentary life, whether there are changes in the daily movement patterns, and more. This thesis describes how various classifiers referred to in the state of the art for the detection of the transport mode have been trained and assessed. The classifier, Decision Tree, showed better performance in the transport mode detection and it was implemented so that it could be used by other Android applications. In order to do it, a system was implemented as proof of concept to demonstrate the operation of the transport classifier. The classifier was incorporated in an application for elderly care, which can be used by health professionals and alike to monitor the daily activity patterns of the elderly. The classifier was created using more than 24 hours of transportation data from a group of 15 individuals and Weka classification shows it can achieve over 96.1% overall accuracy. However, a real world validation of the implemented system obtained 88.97% accuracy. In alignment with the investigation done in this thesis, a paper was submitted and accepted for conference publication at the IEEE HealthCom. The Appendix C presents the accepted paper.
With an increase in average life expectancy, the aging of the population is discernible and consequently a reduction in social and economic conditions for elderly daily care. Taking this into account, more attention is being given to remote care systems to assist the patients and help them take care of themselves, lowering the conventional health care necessity. Human physical activity monitoring is an important feature of such systems, and has received an increasing interest from health-related professionals, such as physiotherapists, elders’ caregivers and nutritionists. This project’s main purposes are the detection of transport modes (walking, bus, car, metro, ...) using data from different sensors available on a common smartphone, and the addition of methods that take advantage of the surrounding environment, such as Wi-Fi Access Points, in order to improve the global accuracy of the system. By monitoring the most common modes of transport used by this population it is possible to support the work of caregivers, giving them knowledge about the user's activity habits, perceive whether the elder is having a sedentary life, whether there are changes in the daily movement patterns, and more. This thesis describes how various classifiers referred to in the state of the art for the detection of the transport mode have been trained and assessed. The classifier, Decision Tree, showed better performance in the transport mode detection and it was implemented so that it could be used by other Android applications. In order to do it, a system was implemented as proof of concept to demonstrate the operation of the transport classifier. The classifier was incorporated in an application for elderly care, which can be used by health professionals and alike to monitor the daily activity patterns of the elderly. The classifier was created using more than 24 hours of transportation data from a group of 15 individuals and Weka classification shows it can achieve over 96.1% overall accuracy. However, a real world validation of the implemented system obtained 88.97% accuracy. In alignment with the investigation done in this thesis, a paper was submitted and accepted for conference publication at the IEEE HealthCom. The Appendix C presents the accepted paper.
Description
Keywords
Reconhecimento de atividades humanas Processamento de sinal Extração de características Machine Learning Cuidado de idosos Deteção de transportes Human Activity Recognition Signal Processing Feature Extraction Machine Learning Elderly care Transport Detection