Hand Angle Estimation Based on sEMG and Inertial Sensor Fusion

Continuous kinematics estimation based systems from surface electromyographic (sEMG) signals can predict movement intention to drive upper-limb prostheses in a more natural and intuitive way. Recently studies about inertial sensors combined with this kind of biosignal have improved the regression algorithms performance significantly. The goal of this work is to propose and evaluate strategies to control prostheses that can be applied in a virtual hand training platform during the training phase of amputee patients. This work presented strategies based on MLP neural networks to estimate 18 DoFs across hand joint angles using sEMG and Inertial Sensor Fusion (ISF). The dataset used was obtained of the seventh Ninapro database. The regression algorithms applied are based on time-domain features, such as auto-regressive model using recursive least squares, slope-sign changes and waveform length. The sEMG and ISF based strategy showed equivalent performance compared to strategy based exclusively on sEMG signals. The results suggest the usability of inertial sensors as a source signal in multi-modal control of upper-limb prostheses.