Features Selection for Force Myography Based Hand Gesture Recognition

Abstract

Hand gesture recognition has a wide range of applications in robotics, game control, and in communication with the deaf and people with trouble hearing. Recognition of American sign language (ASL) hand gestures has been extensively studied in the literature. Multiple data sources and different features extracted from these data were used to classify ASL gestures. In this study, we examined the features used in previous research to determine the minimum number of features that can give an accurate classification of ASL hand gestures. Force myography (FMG) signals captured for ASL gestures of digits 0–9 were used in this analysis of the selected features. Extracted features from the raw FMG signals were applied to K-nearest neighbors (KNN) and Extreme Learning Machine (ELM) to evaluate their efficiency in identifying the corresponding hand gesture. Results show that using only the mean value as input to classification algorithms yields the highest classification accuracy. The classification accuracy was 90% and 96.9% for KNN and ELM, respectively.