Continuous Hand Gesture Semg Recognition Based On Empirical Formula

Most of the studies in the field of hand gesture recognition based on surface electromyography (SEMG) always employ a signal processing scheme used in speech recognition, which includes data measurement, active segmentation, feature extraction and classification. Such framework just works based on the assumption that every hand gesture should be performed from a relaxed state through muscle contraction for a while to the relaxed state again, so that the active segments, which are SEMG signal segments corresponding to the gesture performance, can be easily detected with continuous collected SEMG signal stream. However, the actual hand gesture performance is different from the assumption: actually, a hand gesture is always able to switch to another without a relaxed state between them, and it is not required to exert great strength to keep a hand gesture that has already been performed. Sometimes a hand gesture maintains while relevant muscles stay relaxed. The conditions introduced above are referred to as continuous hand gestures, for which it is impossible to detect exact active segments within SEMG signals.Aiming at the continuous hand gesture recognition based on SEMG, this work tries to meet two major challenges in this field: 1) recognizing multiple consecutive hand gestures without applying active segmentation on the SEMG data; 2) automatically determining the onset and offset of each hand gesture and giving the results instantly rather than after the gesture is complete.In order to realize continuous hand gesture recognition using SEMG, a method for continuous hand gesture recognition based on an empirical formula is proposed in this paper. In this method, the sliding windows are first applied on the raw SEMG data to extract signal energy features within each window. Then, the energy feature is evaluated by well-trained templates of each class. Last, the final decision could be made as the class of which the maximal likelihood is achieved according to the evaluation. With this idea in mind, a formula to describe the probability of testing samples belonging to each hand gesture class is derived from the hand gesture features. The empirical coefficients for the formula are determined by a data processing experiment. The performance of SEMG classification based on our proposed empirical formula is quantified via the experiment on continuous hand gesture recognition. The empirical coefficients for the formula are able to be determined by experimental method effectively, and promising results on the EMG-based continuous hand gesture recognition can be achieved through proposed empirical formula. The experimental results demonstrate the effectiveness of applying such empirical formula on SEMG-based hand gesture recognition. The proposed method using empirical formula provides a practical solution to SEMG-based real-time continuous hand gesture recognition.