Publicación: Exploración de patrones en series temporales de datos inerciales usando técnicas de segmentación no supervisada
Cargando...
Fecha
2025-02
Autores
Editor/a
Director/a
Tutor/a
Coordinador/a
Prologuista
Revisor/a
Ilustrador/a
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Nacional de Educación a Distancia (UNED). E.T.S. de Ingeniería Informática
Resumen
Este Trabajo de Fin de Máster (TFM) explora la aplicación de técnicas de segmentación no supervisada a series temporales de datos inerciales para el análisis de habilidades motoras en Aikido. El TFM se basa en investigaciones previas que emplearon métodos supervisados, buscando validar estos resultados con un enfoque exploratorio. El objetivo principal es identificar patrones y estructuras subyacentes en los datos sin necesidad de etiquetas predefinidas, diferenciándose de enfoques supervisados anteriores. La investigación busca validar si la segmentación no supervisada puede identificar segmentos significativos en series temporales de datos inerciales. Se analizan dos datasets: uno de golpes con espada (Bokken) (D1) y otro de desplazamientos de rodillas (Shikko) (D2). Se realiza una extracción de características tanto de los datos de los participantes, como las series temporales, con el fin de homogeneizar la información, utilizando descriptores estadísticos, temporales y espectrales, y se seleccionan las más relevantes con el modelo LASSO. Para ello se emplean datos de sensores inerciales, específicamente acelerómetros y giroscopios, para capturar movimientos tridimensionales. Los datos provienen de estudios previos de Portaz y Corbí, quienes utilizaron técnicas de clasificación supervisada. En este TFM, se implementan algoritmos como K-Means y K-Medoids. Estos algoritmos se seleccionaron por su capacidad para identificar transiciones y patrones en los datos sin etiquetas. El análisis compara los resultados de los algoritmos no supervisados con segmentaciones manuales y enfoques supervisados, para validar la efectividad de cada método en la clasificación de niveles de experiencia en Aikido, buscando validar y enriquecer estos resultados. Finalmente, se discuten las limitaciones y se proponen futuras líneas de investigación.
This Master's Thesis explores the application of unsupervised segmentation techniques to time series of inertial data for the analysis of motor skills in Aikido. The TFM builds on previous research using supervised methods, seeking to validate these results with an exploratory approach. The main objective is to identify underlying patterns and structures in the data without the need for predefined labels, differing from previous supervised approaches. The research seeks to validate whether unsupervised segmentation can identify meaningful segments in inertial time series data. Two datasets are analysed: one of sword strikes (Bokken) (D1) and one of knee displacements (Shikko) (D2). A feature extraction of both participant data and time series is performed in order to homogenise the information, using statistical, temporal and spectral descriptors, and the most relevant ones are selected with the LASSO model. Data from inertial sensors, specifically accelerometers and gyroscopes, are used to capture three-dimensional movements. The data comes from previous studies by Portaz and Corbí, who used supervised classification techniques. In this TFM, algorithms such as K-Means and K-Medoids are implemented. These algorithms were selected for their ability to identify transitions and patterns in unlabelled data. The analysis compares the results of the unsupervised algorithms with manual segmentations and supervised approaches to validate the effectiveness of each method in the classification of Aikido experience levels, seeking to validate and enrich these results. Finally, limitations are discussed and future lines of research are proposed
This Master's Thesis explores the application of unsupervised segmentation techniques to time series of inertial data for the analysis of motor skills in Aikido. The TFM builds on previous research using supervised methods, seeking to validate these results with an exploratory approach. The main objective is to identify underlying patterns and structures in the data without the need for predefined labels, differing from previous supervised approaches. The research seeks to validate whether unsupervised segmentation can identify meaningful segments in inertial time series data. Two datasets are analysed: one of sword strikes (Bokken) (D1) and one of knee displacements (Shikko) (D2). A feature extraction of both participant data and time series is performed in order to homogenise the information, using statistical, temporal and spectral descriptors, and the most relevant ones are selected with the LASSO model. Data from inertial sensors, specifically accelerometers and gyroscopes, are used to capture three-dimensional movements. The data comes from previous studies by Portaz and Corbí, who used supervised classification techniques. In this TFM, algorithms such as K-Means and K-Medoids are implemented. These algorithms were selected for their ability to identify transitions and patterns in unlabelled data. The analysis compares the results of the unsupervised algorithms with manual segmentations and supervised approaches to validate the effectiveness of each method in the classification of Aikido experience levels, seeking to validate and enrich these results. Finally, limitations are discussed and future lines of research are proposed
Descripción
Categorías UNESCO
Palabras clave
Citación
Ventura Farias, Irene Josefina. Trabajo Fin de Máster: Exploración de patrones en series temporales de datos inerciales usando técnicas de segmentación no supervisada. Universidad Nacional de Educación a Distancia (UNED) 2025
Centro
E.T.S. de Ingeniería Informática
Departamento
Inteligencia Artificial