The Study Of Anthropomorphic Prosthetic Hand Control Based On Electromyography (EMG)

The anthropomorphic prosthetic hand which has multiple degree of freedoms is oftenmounted on an amputee and can be controlled by the Elelctromyography (EMG). Theprosthetic hand can perform some prehensile gestures, such as holding a cup, pinching a keyand so on. However, the control strategy based on the EMG has some weaknesses. It can’trecognize plenty of hand gestures and the recognition accuracy is not very high. This paperimprove the control strategy based on pattern recognition and recognize eight prehensilegestures with a high accuracy. And the control experiments of anthropomorphic virtual handand mechanical hand are implemented.First of all, we review the EMG control strategies in recent years, especially the strategybased on pattern recognition. And our aim is to recognize eight prehensile gestures includingcylindrical, hook, lateral, point, rest, spherical, tripod and tip with only two electrodes. Amyoelectrical control system is designed and it has four parts including signal source, signalacquisition, signal processing and prosthetic hand control. Based on this system, we completethe gestures recognition off-line and om-line, respectively.During the off-line training phase, the Mean Absolute Value, Variance and the forth-orderAutoregressive Coefficient are selected as optimal features and they compose a12dimensionfeature vector. Then the LDA algorithm is used to implemented feature reduction and the bestdimension is defined as7. The current projected feature vector (PFV) and the previous oneare smoothed before the current PFV is sent into the classifier. This step is called“pre-smoothing”. For the on-line test, an over-lapping window scheme is implemented andthe feature Sample Entropy is added into original feature set. The over-lapping windowscheme yields a decision flow. We smooth the current decision and the previous12decisionsto get a final decision and this step can be called “post-smoothing”. The combination ofpre-smoothing and post-smoothing can bring high recognition accuracy and makes therecognition of continuous gestures possible. This paper also builds a virtual hand to displaythe recognition result. And a mechanical hand has been designed, which is expected to carryobjects under the control of EMG signal in the future.