Scholarlink

<p>Traditional observational ergonomics relies on predefined checklists that are time-consuming, subjective, and incapable of capturing the temporal dynamics of manual work. To overcome these limitations, this study proposes an automated, real-time framework that integrates deep learning–based human pose estimation (OpenPose) to automatically extract 2D joint coordinates. Using Python, the system computes biomechanical joint angles from OpenPose-derived keypoints and applies rule-based REBA and RULA protocols to assess posture-specific ergonomic risk in real time. This pipeline enables fully automated, markerless, and real-time posture risk assessment, eliminating the need for wearable sensors, manual annotation, or intrusive instrumentation. The system was validated across five manual handling tasks, including repetitive packaging, bilateral 5&#xa0;kg sachet filling, and overhead carton stacking, with pose estimation accuracy verified against manual Kinovea measurements (MAE &lt; 2.11°, RMSE &lt; 2.75°). Results reveal a wide spectrum of ergonomic risk: REBA scores range from 5.13 (Moderate) to 9.85 (Very High), while RULA scores reach 7.00, with critical segments consistently identified (e.g., left shoulder in overhead work, right forearm in cutting). The framework requires only a standard HD camera and modest computational resources, making it scalable for real-world deployment. By transforming passive video into actionable, posture-specific risk intelligence, this approach enables proactive ergonomic interventions and advances the integration of human-centric digital models in occupational health and 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

Automatic real-time ergonomic posture assessment using digital models: a case study of manual handling tasks

  • 期刊论文

  • 发表日期：

    2025年12月17日

查看全文
  • Nour El Houda Benharkat,
  • Souad Bentaalla Kaced,
  • Ammar Chakhrit,
  • Abdelmalek Chergui

摘要

Traditional observational ergonomics relies on predefined checklists that are time-consuming, subjective, and incapable of capturing the temporal dynamics of manual work. To overcome these limitations, this study proposes an automated, real-time framework that integrates deep learning–based human pose estimation (OpenPose) to automatically extract 2D joint coordinates. Using Python, the system computes biomechanical joint angles from OpenPose-derived keypoints and applies rule-based REBA and RULA protocols to assess posture-specific ergonomic risk in real time. This pipeline enables fully automated, markerless, and real-time posture risk assessment, eliminating the need for wearable sensors, manual annotation, or intrusive instrumentation. The system was validated across five manual handling tasks, including repetitive packaging, bilateral 5 kg sachet filling, and overhead carton stacking, with pose estimation accuracy verified against manual Kinovea measurements (MAE < 2.11°, RMSE < 2.75°). Results reveal a wide spectrum of ergonomic risk: REBA scores range from 5.13 (Moderate) to 9.85 (Very High), while RULA scores reach 7.00, with critical segments consistently identified (e.g., left shoulder in overhead work, right forearm in cutting). The framework requires only a standard HD camera and modest computational resources, making it scalable for real-world deployment. By transforming passive video into actionable, posture-specific risk intelligence, this approach enables proactive ergonomic interventions and advances the integration of human-centric digital models in occupational health and safety.