Aligning Computer Vision with Expert Assessment: An Adaptive Hybrid Framework for Real-Time Fatigue Assessment in...

Area of Science:

• Occupational Health and Safety
• Biomechanical Engineering
• Artificial Intelligence in Healthcare

Background:

• Manual edge-banding in furniture factories leads to high rates of work-related musculoskeletal disorders (WMSDs).
• Existing risk assessment methods suffer from poor cumulative risk quantification and inconsistent evaluations.
• There is a need for continuous, automated, and non-invasive monitoring systems for WMSD risk.

Purpose of the Study:

• To develop and validate a three-stage system for continuous, automated, non-invasive WMSD risk monitoring.
• To improve the accuracy and consistency of WMSD risk assessment and fatigue prediction.
• To provide an engineered solution for WMSD prevention in manual labor environments.

Main Methods:

• Utilized MediaPipe 0.10.11 for extracting 33 key joint coordinates and computing seven joint angles.
• Converted joint angles into graded parameters using RULA, REBA, and OWAS criteria for automatic posture risk scoring.
• Developed an Adaptive Pooling CNN-LSTM hybrid model with ECA for predicting expert-rated fatigue states from nine-dimensional features.

Main Results:

• System-generated posture risk ratings demonstrated strong correlation with expert evaluations, confirming system validity.
• The hybrid CNN-LSTM model outperformed standalone CNN and LSTM models in predicting worker fatigue patterns.
• The system accurately mapped fatigue indexes and generated actionable intervention recommendations.

Conclusions:

• The proposed system offers a valid and reliable method for continuous WMSD risk assessment and fatigue monitoring.
• This automated approach overcomes limitations of traditional manual evaluations, such as subjectivity and poor temporal resolution.
• The study provides a technical reference for occupational health management in labor-intensive industries, enhancing WMSD prevention strategies.