Monitoring indoor human activities for Ambient Assisted Living

At the end of the 20th century, Ubiquitous Computing and Ambient Intelligence were introduced as a vision of the future society. In this context, the paradigm of Ambient Assisted Living (AAL) has allowed the evolution of methods, techniques and systems to improve everyday life, by supporting people in both physical and cognitive aspects, especially in case of the so called “fragile people”. The state-of-the-art research develops means for vital data measurements, for recognizing activities and inferring whether a self-care task has been performed. These results are obtained through the simultaneous presence of different technologies deployed into physical environments in which people live. The monitoring of human activities is fundamental to enable the AAL paradigm. For instance, people spend sleeping several hours a day, thus monitoring this activity is fundamental in understanding and characterizing a person’s sleep habits. On the other hand, at daytime, several indoor activities can be inferred by knowing the exact position of a subject. In this view, the main goal of this thesis is the proposal of advancements in the field of both daytime and night-time monitoring of human activities, focusing on indoor localisation and sleep-monitoring as key enablers for AAL. Regarding Indoor Positioning Systems (IPSs), the lack of a standardized benchmarking tool and of a common and public dataset to test and to compare results of IPSs is still a challenging open issue. Advancements in this direction can lead to improve the performance evaluation of heterogeneous systems, and, consequently, to obtain improvements of the IPSs. Some steps have been made towards introducing benchmarking tools, for example, through the introduction of the EvAAL framework, that defines tool and metrics usable for comparing both real-time and offline methods. This thesis contributes by proposing (i) some improvements to the EvAAL benchmarking framework, especially considering real-time smartphone-based positioning systems; (ii) presenting a common, public, multisource and multivariate dataset, gathered using both a smartwatch and a smartphone, to allow researchers to test their own results. Then, this thesis focuses on both single-device and multiple-device localisation. Concerning single-device positioning strategies, several smartphone-based systems have been recently presented, based on data gathered from smartphone built-in sensors, though with performances not completely satisfactory. In this view, the thesis proposes a novel approach based on deep convolutional neural networks, in order to improve the use of the pedometer (one of the main smartphone built-in sensors used in IPSs) e consequently the Pedestrian Dead Reckoning algorithm performances. Finally, we extend the concept of a single-device localisation to several devices in indoor environments. Localising multiple devices into the same environment can lead to detect, for example, social behaviour and interaction. Several systems try to reach the goal in AAL scenarios but using an intrusive and expensive ad-hoc infrastructure. Instead, we propose a novel approach for finding the presence of people in indoor locations, through a cheap technology as Wi-Fi probes, demonstrating the feasibility of this approach. Regarding the sleep monitoring problem, recent findings show that sleep plays a critical role in reducing the risk of dementia and preserving the cognitive function in old adults. However, state-of-the-art techniques for understanding the sleep characteristics are generally difficult to deploy in an AAL scenario. This suggests that more effort should be spent to find sleep monitoring systems able to detect objective sleep patterns and, at the same time, easy to use in a home setting. In this thesis we propose a system able to perform the human sleep monitoring in an unobstrusive way, using force-sensing resistor sensors placed in a rectangular grid pattern on the slats, below the mattress; it can also detect human bed postures during sleep sessions and to identify patient movements and sleep stages, an information particularly useful, for instance, to assure the pressure ulcer prevention. The proposed advancements have been thoroughly evaluated in the laboratory and in real-world scenarios, demonstrating their effectiveness.