A Single-Camera Method for Estimating Lift Asymmetry Angles Using Deep Learning Computer Vision Algorithms

Area of Science:

• Biomechanics
• Ergonomics
• Computer Vision

Background:

• The revised NIOSH lifting equation is crucial for assessing ergonomic risks.
• Accurate measurement of the asymmetry angle (A) is essential for this assessment.
• Current methods for measuring A can be cumbersome and require specialized equipment.

Purpose of the Study:

• To develop and validate a computer vision (CV) method for automatically measuring the asymmetry angle (A) from a single camera.
• To compare the CV method's performance against 3D motion capture (MoCap) ground truth.

Main Methods:

• A laboratory study with ten participants performing various lifts was conducted.
• A CV method using a 2D pose estimator (HR-Net) and a 3D algorithm (VideoPose3D) was employed.
• Video-derived landmark coordinates were used to estimate A, compared against MoCap data.

Main Results:

• The CV method demonstrated a mean absolute precision error of 6.25° compared to MoCap.
• The mean absolute accuracy error of the CV method against MoCap markers was 9.45°.
• The method effectively estimated A from video data, addressing real-world challenges.

Conclusions:

• The developed CV method provides a viable, automated approach for measuring the asymmetry angle in lifting tasks.
• This technology has the potential to improve ergonomic assessments and reduce workplace injuries.
• Further research can explore its application in diverse occupational settings.