Performance Comparison of Gesture Recognition System Based on Different Classifiers

The hand plays a very important role in our daily life, and the amputees suffer a lot from the loss of hands or upper limbs. Hence, assisting devices are desired urgently. Today, the prosthetic hands based on surface electromyography (sEMG) signals can recognize many hand gestures, but some problems still exist. To identify more gestures, some recognition systems require multiple electrodes, which are unable to be applied to the amputees with less residual muscles. Meanwhile, better computing performance is required as the number of electrodes increases, which is difficult to be applied to the real-time embedded systems. In this article, we aim to recognize six hand gestures by using sEMG sensors as little as possible. To realize this goal, we compare the accuracy and processing time of different feature extraction and classification methods offline, and the results indicate that the combination of time-domain features and backpropagation neural network has better performance. In total, nine subjects participated in the offline experiments, and the accuracy is up to 95.46% by employing two sEMG sensors to recognize six hand gestures.