The Research And Application Of SEMG-based Human Upper Limb Motion Recognition

Surface Electromyograghy(sEMG) signals recorded from interested muscles can be used to decode human's motion intent through suitable feature extraction and motion recognition methods. Although EMG-based motion pattern recognition has been extensively investigated for various purposes, coordinated arm motion involving multiple DoFs across multiple joints has never been estimated from EMG signals. This paper aims to interpret human's arm motion intent only from EMG signals in order to control prosthetic arms or arm exoskeletons as human intends. The main contributions are outlined as follows:1)Suitable EMG features and motion recognition methods were selected through motion recognition performance comparison of different features and motion recognition algorithms in arm motion recognition tasks.2) Three performance indices, motion-select time, motion-recognition rate and motion-completion rate were applied to analyze the online performance of motion recognition, especially the main reasons of the recognition errors.3) The performance of the proposed arm motion recognition method was validated by driving different robotic arms, including a virtual arm developed by virtual reality, a prosthetic arm/hand and an arm exoskeleton developed by the idea of motion synergy. The result of experiment demonstrated that human limb movement was successfully classified by the use of the root mean square and linear discriminant analysis.