Electromyography-Based Hand Gesture Recognition System for Upper Limb Amputees

This article presents a low-power embedded platform that recognizes arm gestures by decoding surface electromyography (EMG) signals of amputees. The system consists of ADS1298 as analog front end (AFE), which can acquire 8-channel EMG signals simultaneously. For the validation of this system in real-time environment, four subjects have been recruited, including one transradial amputee. Six activities have been performed by each subject for the EMG pattern recognition experiment. Two cases have been created for testing data analysis of real-time classification in a DSP processor. Case one (feature extraction in DSP processor and testing off-line) shows the highest classification accuracy of 97.60% and the mean classification accuracy of 95.40%. Case two (training and testing within DSP processor) shows the highest classification accuracy of 97.75% and the mean classification accuracy of 92%. This shows that this system is capable of recognizing user intention in the real-time environment, with above 91% accuracy.