A Case Study: FMG-based Gesture Recognition using High-Density Piezoelectric Electronic Skin and Machine Learning

During muscle contractions, force distributions are generated on muscle surfaces due to muscle activity, which is applicable for control in a Human-Machine Interface. It has been proven that the force distribution from the corresponding body motions can be recorded utilizing so-called Force Myography (FMG). Flexible piezoelectric sensors with attractive sensing properties have been widely used in several areas to detect force variations through wearable devices. In this paper, we developed an FMG armband composed of high-density (24 sensors) piezoelectric electronic skin and multi-channel embedded electronics. The FMG armband was used to recognize eleven hand and wrist gestures performed by able-bodied subjects. To do this, two signal-processing approaches (Front-End Approach (EEA) and Feature-based Approach (FBA)) were developed to process the FMG patterns and extract the proper features. The processed FMG patterns were evaluated and identified by employing various classical machine learning algorithms, and an average gesture recognition accuracy of 98% for wrist gestures was obtained. This paper demonstrates the feasibility of using high-density piezoelectric skin for FMG and leads to alternative methods for gesture recognition in biomedical applications.