Advances in sensing and machine learning technologies give rise to personal health coaches that monitor human activities and support healthy behaviour. In this context, wheelchair users can benefit from such technologies as they often suffer from physical ailments due to insufficient or over activity, prolonged improper posture and lack of postural changes. While sitting posture is a well-studied area, current works focus on the recognition of less frequent postures and miss important postures such as slouching and pelvic postures. The benefits gained from activity tracking also remain restricted to able bodied individuals. In this thesis we demonstrate an end-to-end, multimodal, wheelchair sitting posture and activity monitoring system facilitating just-in-time feedback on postural changes. By using an earable to monitor activity and head posture with a complementary filter and performing classification using Force Sensitive Resistors, we show faster and more precise recognition of relevant postures and activity.

Master Thesis: TU Delft repository