An expert system for detection of key pinch and tripod hand grip based on sEMG signals

Abstract

This work presents a system to recognize key pinch and tripod hand grips using surface electromyography. This system made use of sEMG signals collected from 8 subjects for training and testing, making use of a 4-channel system designed to recognize Key Pinch and Tripod Pinch hand grips. Signals were pre-processed using band pass filter and notch filter, and a time domain feature extraction was performed from these signals, namely Mean Absolute Value, Standard Deviation and Wilson Amplitude. This work implements and compares 5 different classifications techniques namely Random Forest, Decision Tree, Logistic Regression, KNN and Adaboost. Random Forest provided the highest testing accuracy of gesture recognition at 88.68% and a precision of 88.88%. The system correctly recognizes key pinch and tripod hand gestures.