A Rapid Adaptation Approach for Dynamic Air-Writing Recognition Using Wearable Wristbands with Self-Supervised Contrastive Learning

Abstract

Highlights Utilizing self-supervised learning, the proposed wearable wristband with a four-channel sensing array and wireless transmission module is developed for tracking air-writing and dynamic gestures. The model can learn prior features from unlabeled signals of random wrist movements, significantly reducing the dependency on extensive labeled data for training. The wristband system rapidly adapts to multiple scenarios after fine-tuning using few-shot data, enhancing user interaction through natural and intuitive communication.

Keywords

• Wearable wristband
• Self-supervised contrastive learning
• Dynamic gesture
• Air-writing
• Human–machine interaction