Non-periodic lower-limb motion recognition with noncontact capacitive sensing

Noncontact captive sensing is a new sensing strategy that we proposed previously to compensate for the limitations of existing surface electromyography studies for exoskeleton control. It has been validated on locomotion mode recognition and gait phase estimation. However, our previous studies addressed the tasks of periodic ambulation based on machine learning algorithms. The performances of the capacitive sensing on non-periodical lower-limb motion recognition have never been evaluated. In this preliminary study, we designed a motion recognition method by fusing the capacitive sensing and the inertial sensors. The recognition algorithm was designed based on the combined logics, which freed the system from burdensome training procedures in the recognition tasks. The method was validated on three healthy subjects in performing 6 lower-limb motions and the transitions between them (10 in total). The capacitance-inertial fusion method produced an average precision/recall of >0.92 in static motions and >0.86 with transitions. The most prominent improvement of using the capacitance signals is that it increases the time-response ability during the motion transitions. Compared with purely using inertial sensors, the sensor fusion method reduced more than 100-ms latency on average. The pilot study extends the scope of the new sensing method in human motion recognition. Future efforts will be paid in this direction to get more promising results.