A Deep-Learning Based Posture Detection System for Preventing Telework-Related Musculoskeletal Disorders

Area of Science:

• Occupational Health
• Computer Vision
• Ergonomics

Background:

• The COVID-19 pandemic shifted many jobs to teleworking, increasing computer use and the risk of poor workstation ergonomics.
• Many home workstations lack proper ergonomic setups, leading to uncomfortable and incorrect worker postures.
• Occupational health professionals require automated tools to assess and quantify poor postural habits.

Purpose of the Study:

• To design, implement, and test an automated system for detecting and quantifying incorrect worker posture.
• To provide real-time feedback and recommendations to workers to prevent musculoskeletal problems.
• To address the need for objective postural habit assessment in teleworking environments.

Main Methods:

• A specialized hardware system was developed for real-time video processing.
• Convolutional neural networks (CNNs) were employed for posture detection.
• The system analyzes the posture of the neck, shoulders, and arms.

Main Results:

• The system achieves real-time video processing at up to 25 frames per second.
• It operates with low power consumption (under 10 watts) on specialized hardware.
• The system demonstrates over 80% accuracy in detecting postural patterns.

Conclusions:

• The developed system effectively detects incorrect worker posture in real-time using CNNs.
• It offers a viable solution for occupational risk prevention in teleworking settings.
• The system's efficiency and accuracy support its use in promoting healthier work habits.