A Comprehensive Survey on EMG-Based Real-Time Gesture Recognition for Prosthetic Hand Applications

Abstract

Electromyographic signal processing for gesture recognition represents the backbone of modern assistive technologies,
prosthetic control, and human–computer interactionsystems. However, high classification accuracy combined withcomputational efficiency i s s till a k ey c hallenge due to noise, motion artifacts, muscle cross-talk, and intersubject variability inherent in sEMG signals. Furthering prior work, this paper investigates an optimized EMG pattern recognition framework
that embeds validated preprocessing techniques, namely bandpass filtering, wavelet-based d enoising, a nd a rtifact suppression
to enhance signal quality before analysis. The system considerslightweight machine learning algorithms involving support vector
machine, K-nearest neighbors, Random Forest, and LDA togetherwith deep learning models such as CNNs and LSTM-based
recurrent networks, which have always reported state-of-the-artperformance in EMG gesture recognition. Experimental validation
on benchmark sEMG datasets evidences accuracies above97%, very well aligned with recent CNN/RNN-based literature
while keeping computational complexity as low as to fit embedded platforms. Variability analysis in terms of electrode placement,
muscle fatigue, and cross-user settings further validates the robustness and reliability of the proposed approach. The novelty
of this paper hence hinges on providing a comprehensive system framework for a reliable real-time implementation of wearable
rehabilitation platforms and devices that communicate using human-friendly gesture-based commands.