Scholarlink

<p>Gesture recognition is a fundamental technology in nearly all Human-Computer Interaction (HCI) devices, drawing significant attention from engineers and researchers worldwide. Fine-grained air-writing recognition within a small spatial range remains a challenging yet crucial research topic. In this paper, we propose a method, namely RingRT, which uses an inertial measurement unit (IMU) integrated in the ring to identify character air-writing actions within an area as big as the palm of one’s hand. RingRT leverages the neural network that integrates both the Transformer architecture and the Focused Linear Attention mechanism, enabling it to effectively analyze 6-axis motion signals and accurately recognize air-written characters. In experiments, we collected 24,120 air-writing samples within a 6&#xa0;cm <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> 6&#xa0;cm area, comprising 67 characters from 24 participants across three scenarios. RingRT achieved a user-independent accuracy of 90.08% in offline evaluation and maintained over 86% accuracy in real-world scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

RingRT: recognizing air-writing on the palm through an IMU-integrated ring

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月30日

查看全文
  • Bohua Feng,
  • Guancheng Chen,
  • Huihui Chen,
  • Aiguo Wang

摘要

Gesture recognition is a fundamental technology in nearly all Human-Computer Interaction (HCI) devices, drawing significant attention from engineers and researchers worldwide. Fine-grained air-writing recognition within a small spatial range remains a challenging yet crucial research topic. In this paper, we propose a method, namely RingRT, which uses an inertial measurement unit (IMU) integrated in the ring to identify character air-writing actions within an area as big as the palm of one’s hand. RingRT leverages the neural network that integrates both the Transformer architecture and the Focused Linear Attention mechanism, enabling it to effectively analyze 6-axis motion signals and accurately recognize air-written characters. In experiments, we collected 24,120 air-writing samples within a 6 cm \(\times\) × 6 cm area, comprising 67 characters from 24 participants across three scenarios. RingRT achieved a user-independent accuracy of 90.08% in offline evaluation and maintained over 86% accuracy in real-world scenarios.