A hierarchical approach to recognize purposeful movements using inertial sensors: preliminary experiments and results

Abstract

One of the most relevant post-stroke conditions is the hemiparesis, which causes muscle weakness and/or the inability to move one side of the body. Physical and occupational therapy plays an important role in the rehabilitation of patients suffering this condition. On the other hand, daily life use of the impaired arm is crucial for improving and also assessing the evolution of the patient. Currently, this assessment is done through self-questionnaires and interviews, which are subjective and depend on the memory of the patient. In this paper, a hierarchical automatic approach aimed at recognizing purposeful arm movements during patients' daily life activities is presented. This approach relies on two-levels: the former is aimed at distinguishing between arm movement and non-movement; whereas the latter is devoted to recognize between purposeful and non-purposeful movements. In particular, in the first version of the system, we consider arms swing while walking as non-purposeful movement. Experiments have been performed in the lab with 9 healthy volunteers wearing a wristband on each wrist. Six activities have been performed: eating, pouring water, drinking, brushing their teeth, folding a towel, and walking. The proposed approach achieves promising performances, recognizing purposeful movement with an accuracy of 0.91 and an F1-score of 0.87.