Álvarez de la Concepción, Miguel Angel

12 de Septiembre de 2018

  • Autor: Álvarez de la Concepción, Miguel Angel.
  • Titulo: “Desarrollo y Versatilidad del Algoritmo de Discretización Ameva”
  • Departamento: Lenguajes y Sistemas Informáticos.
  • Teseo: https://www.educacion.gob.es/teseo/mostrarRef.do?ref=1675902
  • Directores: Juan Antonio Álvarez García, Luis Miguel Soria Morillo y Juan Antonio Ortega Ramírez (Tutor/Ponente).
  • Sinopsis:

    This thesis, presented as a set of research papers, studies the problem of activity recognition and fall detection in mobile systems where the battery draining and the accuracy are the main areas of researching. These problems are tackled through the establishment of a new selection, discretization and classification algorithm based on the core of the algorithm Ameva. Thanks to the discretization process, it allows to get an efficient system in terms of energy and accuracy.

    The new activity recognition algorithm has been designed to be run in mobile systems, smartphones, where the energy consumption is the most important feature to take into account. Also, the algorithm had to be efficient in terms of accuracy giving an output in real time. These features were tested both in a wide range of mobile devices by applying usage data from recognized databases and in some real scenarios like the EvAAL competition where non-related people carried a smart-phone with the developed system. In general, it had therefore been possible to achieve a trade-off between accuracy and energy consumption.

    The developed algorithm was presented in the Activity Recognition track of the competition EvAAL (Evaluation of Ambient Assisted Living Systems through Competitive Benchmarking), which has as main objective the measurement of hardware and software performance. The system was capable of detecting some activities through the established set of benchmarks and evaluation metrics. It has been developed for multi-class datasets and obtains a good accuracy when there is approximately the same number of examples for each class during the training phase. The solution achieved the first award in 2012 competition and the third award in 2013 edition.