No abstract

Control of human hand using surface electromyography (EMG) is already established in various mechanisms, but proportionally controlling magnitudes degrees of freedom (DOF) of humanoid hand model is still highly developed in recent years. This paper proposes another method to achieve a proportional estimation and control of human’s hand multiple DOFs. Gestures in the form of American Sign Language (ABCDFIKLOW) were chosen as the targets, of which ten alphabetical gestures were specifically used following their clarity on its 3D model. Then the dataset of the movements gestures was simultaneously recorded using High-density electromyography (HD-EMG) and motion capture system. Sensor placements were on intrinsic - extrinsic muscles for HD-EMG and finger joints for the motion capture system. To derive the proportional control in time series between both datasets (HD-EMG and kinematics data), neural network (NN) and k-Nearest Neighbour were used. The models produced around 70-95 % (R index) accuracy for the eleven DOFs in four healthy subjects’ hand. kNN’s performance was better than NN, even if the input features were reduced either using manual selections or principal component analysis (PCA). The time series controls could also identify most sign language gestures (9 of 10), with difficulty was given on O gesture. The false interpretation was because of nearly identical muscle’s EMG and kinematics data between O and C. This paper intends to extend its conference version [1] by adding more in-depth Results and Discussion along making other sections more comprehensive.