Development of smart multi-sensor solutions for the Active Aging and Well Being

The European population aging poses various challenges, especially in the health care scenario. This situation has been perceived by many as a threat to Europe's economy and competitiveness, in particular when it comes to the sustainability of the healthcare systems. From a technical point of view, the greatest risks associated with a progressive increase of the population frailty, concern the onset of falls, often associated or preceded by postural instability, and the loss/reduction of autonomy. Often, these phenomena are further stressed by wrong habits in terms of nutrition and mobility. These are then addressed problems within this PhD thesis. In particular, the applications addressed by the following thesis are: detection and classification of falls, analysis and classification of postural instabilities, analysis and detection of user's habits. These three applications are intrinsically connected. Concerning falls, an event-driven methodology, based on template matching techniques is proposed, analyzed, and validated where the adopted metrics and related classifiers, were selected to reduce the algorithm computational complexity. One of the main objectives, common to all the presented activities, is the development of methodologies that can be easily integrated into low power microcontroller platforms. The results obtained in terms of classification between fall and non-fall events, as much as the classification among the specific falls, allow us to quantify the goodness of the proposed solution, also by comparing it with more complex methodologies present in the state of the art. In the context of postural instabilities, the use of inertial platforms has been validated against clinical solutions, such as force platforms and vision systems, through a careful analysis of results obtained by the comparison between the proposed solution and a reference system. Furthermore, in addition to the metrics qualifying the ability of the classifier to predict whether a given event belongs to the correct class, a reliability index has been defined and used to quantify the quality of the individual classification process (i.e. how much you need to trust that specific classification). The results obtained demonstrate the feasibility of using a low-cost inertial platform for the continuous analysis of postural conditions, even during normal daily operations. A further proposal, in the context of instabilities, has been based on the use of the Wavelet transform. This particular mathematical transformation has been widely used in the literature for the analysis of the postural conditions of the subjects, with specific reference to the evaluation of posture control mechanisms. However, it should be specified that the results obtained from the analysis are rarely used to classify postural conditions. For this reason, the proposed methodology makes use of the Wavelet transform, in particular the discrete Wavelet transform, on which output features are calculated. A detailed analysis was made on the contribution made by each individual feature and, based on obtained results, it was possible to define the specifications of the needed classifier. In particular, a K-Nearest Neighbor classifier was employed. This choice was based on specific properties that make the classifier easily embeddable into microcontroller platforms. Obtained results, in terms of Sensitivity and Specificity, show the validity of the proposed solution. A further application, concerning the development and characterization of a device for user' habits monitoring based on RFID technology, is presented. In particular, the device was developed within an Interreg project (NATIFLife) which aimed at the development of an innovative framework of assistive devices that could improve elderly autonomy. In detail, the project aims at the assessment of the user habits, activity rate, nutrition and hydration, as well as the use of home appliances. Conducted experiments were mainly aimed at the optimization of the system characteristics, in terms of physical dimension, power consumption, and reading range. Moreover, to avoid miss-identification, resulting in a miss-confidence of the solution, an assessment procedure is conducted aiming at evaluating the system reliability when used in a scenario simulating a domestic area. The adopted metrics have proved the robustness and reliability of the proposed solution. All the activities, therefore, have a single common goal: improving the living conditions of fragile subjects, increasing self-confidence and relative autonomy.