ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-09102010-102350


Tipo di tesi
Tesi di laurea specialistica
Autore
BONATESTA, FRANCESCO
URN
etd-09102010-102350
Titolo
Fall Detection: A Self-Learning Ubiquitous System
Dipartimento
INGEGNERIA
Corso di studi
INGEGNERIA INFORMATICA
Relatori
relatore Prof. Corsini, Paolo
relatore Dott. Vecchio, Alessio
relatore Prof. Avvenuti, Marco
Parole chiave
  • health care
  • mobile phone
  • fall detection
  • neural network
  • accelerometer sensor
Data inizio appello
15/10/2010
Consultabilità
Non consultabile
Data di rilascio
15/10/2050
Riassunto
Today's life expectancy is becoming higher and higher thanks to the development of medical technologies and increased quality of life. One major factor that influence the physical health of a person is a fall event, which also has also dramatic psychological consequences, since it drastically reduces the self-confidence and independence of the affected people.

Even if a fall is hard to predict, it is of fundamental importance to provide health-care to the injured people as soon as possible, especially when she is not able to just "press the button". The consequences of a fall event are, in fact, directly proportional to the long-lie period (i.e. the time interval during which the person remains involuntarily on the ground) after the fall.

In this work, we explore different techniques in the area of wearable devices, with the aim of designing an automatic fall detection system which can be effectively usable by the vast majority of potentially interested users in most situations. In this sense we propose a solution which integrates wearable sensors with mobile phones, which all people are already familiar with. Also, this kind of solution eliminates the need for any other external infrastructure and provides new opportunities to exploit smartphones' sensing and communication capabilities.

In particular, in our solution we employ a single accelerometer sensor to be placed at the user's waist connected to her mobile phone through the Bluetooth technology, with an option to use only the mobile phone if it embeds a sensor suitable for our purposes. We have included the possibility (for users) to deactivate the system in case of false alarm and an option for the user to provide hints and suggestions to the system.

Thanks to the added interactivity, we investigated the applicability of a machine learning approach and proposed a system made of a feature extractor built starting from domain knowledge and a neural network which acts as a pattern classifier.

Finally we analyzed the results of a usability questionnaire that has been given to a number of potential users of the system to examine their reactions and feelings about our proposed solution.
File