Hand motion recognition with postural changes using surface EMG signals

In this paper, we consider a hand motion recognition problem using surface Electromyography signals (EMGs). Most previous studies commonly assume that the relationship between the EMG signal (or feature) and the motion intention is invariant. However, such an assumption cannot be satisfied for hand motion recognition if the user changes the posture of the arm (e.g., pronation angle) that affects on the relative positions of the sensors from the target muscles. We propose a robust motion classifier for such a postural change using pattern matching techniques. The effectiveness of our proposed method is validated by experiments with five subjects.