| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 13 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O envelhecimento demográfico trouxe novos desafios à sociedade. Associado à prevalência do risco de quedas na população idosa, são primordiais projetos de desenvolvimento que sejam uma mais valia na prevenção desta realidade e que minimizem os seus danos. Contudo, antecipar ou mesmo detetar uma queda humana é uma tarefa complexa, devido a limitações funcionais relacionadas com a condição de dependência do utilizador, a dificuldade em distinguir o evento de atividades do dia a dia, ou constragimentos que afetem a privacidade da pessoa. Relevando o privilégio de contribuir para a manutenção da qualidade de vida dos idosos, pretende-se com esta dissertação efetuar um estudo detalhado sobre dispositivos e sensores de deteção de quedas, baseado nas vibrações e áudio do piso da divisão utilzada pela pessoa idosa. Após este estudo, será apresentada uma solução cuja arquitetura de computação envolva conceitos como edge computing, fog computing e cloud computing. Neste contexto, a solução proposta integra se no ambiente residencial do idoso, consistindo num dispositivo por divisão e um dispositivo central na habitação. Para o efeito procedeu se ao desenvolvimento do dispositivo IoT (internet of things) a instalar na divisão, cuja função principal será monitorizar os sinais de vibração, som e movimento a partir da leitura dos sensores, sem requerer a manutenção ou configuração por parte do utilizador. Para além das funcionalidades de comunicação ao dispositivo no edifício, o dispositivo na divisão tomará a decisão de um alerta imediato para o cuidador residente, no caso da existência de uma queda e o envio da série temporal para classificação. O dispositivo implementado no edifício realiza a recolha de dados que depois de preprocessados constituem uma base de conhecimento. Ao aplicar a transformada de wavelet e dynamic time warping (DTW) na análise de séries temporais correspondentes ao registo de simulações de quedas e atividades diárias, foi possível efetuar a classificação inteligente da série temporal baseada em algoritmos de aprendizagem. A solução proposta integra-se nas áreas de smart homes e assisted living, usando técnicas de deep learning de forma a identificar e distinguir com precisão uma queda.
Demographic aging has brought new challenges to society. Associated with the prevalence of the risk of falls in the elderly population, development projects that are an asset in preventing this reality and that minimize their damage are essential. However, anticipating or even detecting a human fall is a complex task, due to functional limitations related to the user’s condition of dependence, the difficulty in distinguishing the event from daily activities, or constraints that affect the privacy of the person. Relevant to the privilege of contributing to the maintenance of the quality of life of the elderly, it is intended with this dissertation to carry out a detailed study on devices and sensors for fall detection, based on vibrations and audio of the floor of the division used by the elderly. After this study, a solution will be presented whose computing architecture involves concepts, such as edge computing, fog computing, and cloud computing. In this context, the proposed solution is inserted into the residential environment of the elderly, consisting of a device per division and a central device in the building. For this purpose, the IoT (internet of things) device to be installed in the room was developed, whose main function will be to monitor the vibration, sound, and movement signals from the reading of the sensors, without requiring maintenance or configuration by the user. In addition to the communication features of the device in the building, the device in the room will decide an immediate alert for the resident caregiver, in the event of a fall and the sending of the time series for classification. The device implemented in the building performs the collection of data, which after preprocessing, constitutes a knowledge base. By applying the wavelet transform and dynamic time warping (DTW) in the analysis of time series corresponding to the registration of simulations of falls and daily activities, it was possible to perform the intelligent classification of the time series based on learning algorithms. The proposed solution is integrated into the areas of smart homes and assisted living, using deep learning techniques to accurately identify and distinguish a fall.
Demographic aging has brought new challenges to society. Associated with the prevalence of the risk of falls in the elderly population, development projects that are an asset in preventing this reality and that minimize their damage are essential. However, anticipating or even detecting a human fall is a complex task, due to functional limitations related to the user’s condition of dependence, the difficulty in distinguishing the event from daily activities, or constraints that affect the privacy of the person. Relevant to the privilege of contributing to the maintenance of the quality of life of the elderly, it is intended with this dissertation to carry out a detailed study on devices and sensors for fall detection, based on vibrations and audio of the floor of the division used by the elderly. After this study, a solution will be presented whose computing architecture involves concepts, such as edge computing, fog computing, and cloud computing. In this context, the proposed solution is inserted into the residential environment of the elderly, consisting of a device per division and a central device in the building. For this purpose, the IoT (internet of things) device to be installed in the room was developed, whose main function will be to monitor the vibration, sound, and movement signals from the reading of the sensors, without requiring maintenance or configuration by the user. In addition to the communication features of the device in the building, the device in the room will decide an immediate alert for the resident caregiver, in the event of a fall and the sending of the time series for classification. The device implemented in the building performs the collection of data, which after preprocessing, constitutes a knowledge base. By applying the wavelet transform and dynamic time warping (DTW) in the analysis of time series corresponding to the registration of simulations of falls and daily activities, it was possible to perform the intelligent classification of the time series based on learning algorithms. The proposed solution is integrated into the areas of smart homes and assisted living, using deep learning techniques to accurately identify and distinguish a fall.
Description
Keywords
Classificação de quedas Medição de vibrações Séries temporais Sistemas de aprendizagem Classification of falls Learning systems Times series Vibration measurement