Selective Linear-Regression Model for hand posture discrimination and grip force estimation using surface electromyogram signals

This paper proposes the method of hand posture discrimination and grip force estimation by means of Selective Linear-Regression Model. Generally, myoelectric hands which discriminate hand posture and estimate grip force at the same time result in unsatisfying results because of complication of EMG signals. Therefore, most of myoelectric hands can control either the force or the posture. However, the proposed method is able to discriminate hand posture and to estimate grip force simultaneously while the accuracy results are achieved. In experiments, EMG signals were measured while hand posture and grip force were changing. As a result, it appears that EMG features increase monotonically with grip force. In addition, increasing forms of EMG features are different on each posture. Based on these experimental results, the authors propose the method for both discriminating hand posture and estimating grip force by means of several linear-regression models which utilize the relationship between the grip force and EMG features on each posture. To evaluate the effectiveness of this method, the failure rates of discrimination and the estimation errors of the proposed method were employed. The results indicate that failure rates and estimation errors are improved significantly.