Hand gesture recognition for the control of an exoskeleton

Despite the existence of many examples in multifunctional control systems there is a lack of studies that show hand gestures applied in daily life activities. Furthermore, isometric contractions, above certain thresholds, continue to be used once it is easier to deal with. However, it is a static contraction, that is not used to perform movements. Thus, a control system based on that is not intuitive, especially for subjects who have the limb with a diminished strength and that could also be benefited by rehabilitation devices, such as exoskeletons to train or regain function. In this context, the purpose of this work is to investigate the recognition of up to 4 hand gestures plus the neutral hand position, based on myoelectric signal obtained during the static phase at the end of the movement, without the use of any additional isometric contraction. Performance evaluation is done based on Linear Discriminant Analysis comparing the results of six myoelectric features and also the number of muscles necessary to achieve the best classification accuracy. The results show higher rates for the features in the frequency domain. The Spectral Magnitude Average reaches an average accuracy of 88.44% following by Spectral Moments with 85.56%. The best results achieved by each subject is variable, with a predominant use of 3 to 5 muscles depending on the feature that was used, with no standard pattern.