Scholarlink

Background <p>Ensuring adherence to Standard Operating Procedures (SOPs) is paramount in manufacturing for maintaining product quality, optimizing production efficiency, and safeguarding operator well-being. While current motion tracking techniques often focus on macro movements, the critical details embedded within micro motions are frequently overlooked, hindering a comprehensive understanding of operator performance.</p> Questions/purpose <p>To address this significant limitation, our study introduces a decentralized deep-learning system for real-time operator guidance.</p> Methods <p>By uniquely capturing both the broader macro movements and the subtle yet crucial micro motions simultaneously during SOP execution, our system provides a more granular and effective approach to operator support. Leveraging a novel distributed architecture with specialized deep-learning modules on distinct devices, our system integrates an intelligent flexible operating guidance module powered by YOLO for macro motion detection and a sophisticated kinematic motion analysis module utilizing MediaPipe for precise micro motion recognition.</p> Results <p>In a complex assembly scenario, the system demonstrated compelling performance, achieving 96.67% accuracy, 96.67% recall, and an exceptional 98.2% F1-score.</p> Conclusions <p>The key contribution of this research lies in the novel application of a distributed architecture to enhance the system’s real-time responsiveness and facilitate seamless collaboration, ultimately promising substantial improvements in operational efficiency and a heightened focus on operator safety.</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

Decentralized motion recognition system for operator guidance – a deep learning approach

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Kung-Jeng Wang,
  • Chen-Hsuan Yang

摘要

Background

Ensuring adherence to Standard Operating Procedures (SOPs) is paramount in manufacturing for maintaining product quality, optimizing production efficiency, and safeguarding operator well-being. While current motion tracking techniques often focus on macro movements, the critical details embedded within micro motions are frequently overlooked, hindering a comprehensive understanding of operator performance.

Questions/purpose

To address this significant limitation, our study introduces a decentralized deep-learning system for real-time operator guidance.

Methods

By uniquely capturing both the broader macro movements and the subtle yet crucial micro motions simultaneously during SOP execution, our system provides a more granular and effective approach to operator support. Leveraging a novel distributed architecture with specialized deep-learning modules on distinct devices, our system integrates an intelligent flexible operating guidance module powered by YOLO for macro motion detection and a sophisticated kinematic motion analysis module utilizing MediaPipe for precise micro motion recognition.

Results

In a complex assembly scenario, the system demonstrated compelling performance, achieving 96.67% accuracy, 96.67% recall, and an exceptional 98.2% F1-score.

Conclusions

The key contribution of this research lies in the novel application of a distributed architecture to enhance the system’s real-time responsiveness and facilitate seamless collaboration, ultimately promising substantial improvements in operational efficiency and a heightened focus on operator safety.