Development of a multi-modal system to provide e-health related services based on indoor positioning
- Autores: Gustavo Adolfo Hernandez Peñaloza
- Directores de la Tesis: Federico Alvarez García (dir. tes.)
- Lectura: En la Universidad Politécnica de Madrid ( España ) en 2019
- Idioma: español
- Tribunal Calificador de la Tesis: José Manuel Menéndez García (presid.), Silvia Alba Uribe Mayoral (secret.), Mirela Popa (voc.), Gregorio Ignacio López López (voc.), Ignacio Parra Alonso (voc.)
- Programa de doctorado: Programa de Doctorado en Tecnologías y Sistemas de Comunicaciones por la Universidad Politécnica de Madrid
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
This doctoral thesis is devoted to the modelling of sustainable Integrated Healthcare systems based on the use of Information and Communication Technologies (ICT) tools in combination with advanced processing algorithms. Research is focused on applying novel classification and prediction techniques to information collected from multiple sources. These algorithms jointly form fused models aimed at providing personalized care services to elderly patients.
Multiple modalities of information are understood as all the measurements that can be collected from diverse physical sensors. As an example, person movements can be captured by visual-based sensors (RGB and / or RGB-D cameras), however, it can be fused with inertial information extracted from smart-bands, or with information retrieved from Wireless Sensor Networks (WSN) Devices, or even, binary sensors conveniently allocated. The development of algorithms to exploit similarities yields to significant improvements in the detection/prediction of particular events associated to target groups.
With regard to target groups, this thesis is mainly centered on Parkinson Disease Patients, and therefore in the development of algorithms to detect relevant events associated to such disease. For this ambition, a set of features and descriptors are defined, extracted, pre-processed and modelled. The models created are built on both traditional statistical estimation techniques and Deep Learning techniques.
By significant features, research is sharpened in the extraction of data patterns that have a larger impact in the detection task, as well as the proper weighting strategies for all features available to attain a better performance in terms of accuracy, reducing the time-computing and better adapting to significant changes in the status of an environment.
Finally, extrapolation and spreading of the algorithms scope is reached by its application to diverse fields. The employment of such concepts / architectures for diverse tasks such as genre classification of movies, showing its good performance when compared to traditional visual-based descriptors methods.
Moreover, methods presented in this thesis are also able to increase the efficiency of Generative Model Systems by speeding up the training process. Therefore, the toolbox of algorithms, methods and models proposed contribute to conceive modular systems, specially dedicated for elderly healthcare but generally extendable to other fields.
The hypothesis formulated in this thesis were validated by conducting multiple experiments with end users, creating datasets that can be interesting for research community. Due to the multi-disciplinary environment where this thesis took part, methodological, social and perceptive aspects have been considered. Consequently, some tests performed were evaluated via quantitative (accuracy, recall) and qualitative manners (acceptance, usefulness).
