Recognizing human behaviour using information from Smart Environments

Ambient Intelligence is a new research line in Artificial Intelligence field. Under this paradigm, users interact with an environment that is equipped with different kinds of sensors and actuators. Thanks to those devices, applications in AmI collect users¿ activities and exploit that information, in order to learn users¿ activities and to be able to anticipate their needs. In this thesis, we present a method to understand user daily activities in order to help people to improve their quality of life. Our model increases people¿s autonomy by means of supervising their independent living, without neglecting their safety. With that aim in mind, we propose a system to recognize human behaviour in Smart Environments. Concretely, we model human activities in function of information collected from specific known-environments. That collected information is used to develop algorithms and structures that enable us to follow current user activities and detect anomaly or abnormal behaviour. We focus on the fact that a specific type of behaviour is characterized by a set of common actions, which we assume are the most frequent actions performed during the behaviour execution time. However, as we are dealing with time information, we ascertain that behaviour is not precise or constant, so that, we develop methods to manage the uncertainty and use it to improve the quality of the knowledge. Additionally, we generalize the problem from two points of view: on one hand, we adapt the knowledge to new circumstances, for example, user starts waking up later or besides making coffee, s/he starts having an orange juice every morning. These adaptations provide a tendency, which is studied to extract understandable information. Finally, we generalize the concept of behaviour to manage sets of behaviours, named routines. All the designed models have been tested in Real Ambient Intelligence Scenarios, such as, RFID laboratories, Elderly houses or Intelligent Spaces (iSpace).Tesis Univ. Granada. Departamento de Ciencias de la Computación e Inteligencia ArtificialEsta tesis doctoral ha sido subvencionada bajo el Programa de Becas de Formación del Profesorado Universitario, en la Resolución del 5 de julio de 2008, bajo la referencia AP2007-03578. También ha sido parcialmente financiada con los fondos asociados al proyecto TIN2009-14538-C02-01, de la convocatoria de Proyectos del Plan Nacional, convocatoria 2009, del Ministerio de Ciencia e Innovación