Identificação de gestos por sinais eletromiográficos de superfície utilizando aprendizado de máquina

Abstract

The human-machine interfaces limitations should be constantly challenged, só qe can achieve enhancements or even reestabilsh human capacities. Several topics are studied around this thematic axis, one of then being gesture recognition, that brings an intuitive and pratical control for different systems. The sEMG technique is highlighted in gesture recognition applications, showing high reliability and robustness when the correct signal processing and acquision are executed. In this project it was developed a sEMG signal processing system for hand gesture recognition. Five hand gestures from five subjects were acquired by a commercial Myo™armband, that has 8 channels of dry electrodes. After data acquisition it was evaluated threshold and acquisition windows for the segmentation step, feature groups for the feature extration step and classifier parameters, selecting the best accuracy for usage in online operation. The training strategies of the algorithms are compared between all data collected and subject to subject, also comparing their accuracy. The system developed attained 95.9% accuracy in offline operation and 93.6% in online operation tests, using a SVM classifier.