Demonstration of Cross-Subject Hand Gesture Recognition from EMG

Martin Colot, Cédric Simar, Ana Maria Cebolla Alvarez, Gianluca Bontempi

Track: Type C (Demonstration Abstracts)

Keywords: Human-Computer Interface, EMG, Hand Gesture Recognition, Cross-Subject, Unsupervised Domain Adaptation

Abstract: Electromyography (EMG) enables intuitive human-computer interaction through hand gesture recognition. However, model performance typically drops in cross-subject scenarios due to physiological and session variability. We demonstrate LDA-KM-DA, a novel unsupervised domain adaptation method that adapts to new users without labeled calibration, reaching better accuracy than state-of-the-art methods across datasets and during continuous adaptation. A live demonstration with an 8-channel EMG armband showcases real-time gesture recognition, highlighting its potential for assistive technologies, gaming, and prosthetic devices.

Serve As Reviewer: ~Martin_Colot1

Submission Number: 65