Riconoscimento automatico del cammino in soggetti anziani tramite sensori inerziali in condizioni di vita reale

Gait analysis is the subject of many research projects in the medical field since an accurate and reliable evaluation of gait characteristics can provide key information about people’s health and can predict the onset of several neurodegenerative diseases affecting mainly the elderly population. The purpose of this thesis is to automatically identify gait in elderly population through methods of automatic recognition and to describe gait quality with parameters derived from data acquired through wearable sensors in free-living conditions (where the subjects are free to walk in a natural way without any supervision). First, a gait classification (gait vs other activities) was performed with three distinct approaches: machine learning, deep learning and Gait Event Detection Method (GEDM). Three different sensor solutions (single sensor on L5, single sensor on wrist, two-sensors L5+wrist) were tested for machine learning and deep learning algorithms, while the GEDM method was used only with a single sensor on the lower back (L5). The findings show that the best performance (highest F-measure value) was achieved with the SVM machine learning algorithm. Between the two single sensor solutions the best performance was obtained by the sensor at the lower back. The results for the two-sensor solution are comparable to those obtained on L5. Second, gait analysis was carried out for GEDM and for the best classifier (both L5 and wrist). At the end of the analysis a report file showing gait characteristics was automatically generated; it can aid the clinician in his healthcare task and so contribute to improve wellbeing and quality of life.