Pose estimation with a Kinect for ergonomic studies: evaluation of the accuracy using a virtual mannequin

Area of Science:

• Ergonomics and Human-Computer Interaction
• Biomechanical Engineering
• Motion Capture Technology

Background:

• Kinect-based human pose analysis offers potential for evaluating workstation risks of musculoskeletal disorders.
• Complex 3D poses and environmental constraints in real-world settings challenge the reliability of kinematic data acquisition.
• Predicting measurement accuracy for diverse poses and sensor placements is difficult with traditional experimental methods.

Purpose of the Study:

• To introduce a novel evaluation method using a virtual mannequin for assessing Kinect's human pose measurement accuracy.
• To evaluate joint positions, angles, and Rapid Upper Limb Assessment (RULA) scores across numerous poses and sensor configurations.
• To identify specific scenarios where Kinect's accuracy for ergonomic assessments may be compromised.

Main Methods:

• Development and application of a virtual mannequin-based evaluation framework.
• Automated testing of over 500,000 unique configurations of upper-limb poses and Kinect sensor placements.
• Validation of the virtual mannequin approach against real subject data for a subset of configurations.

Main Results:

• Kinect kinematic data generally demonstrates sufficient accuracy for populating ergonomic assessment grids like RULA.
• Significant increases in inaccuracy were observed for specific, challenging poses and sensor positions.
• The study successfully identified configurations leading to high measurement inaccuracies.

Conclusions:

• The virtual mannequin method provides a scalable and efficient way to evaluate Kinect's accuracy in ergonomic assessments.
• Designers can utilize the supplementary software tool to anticipate sensor accuracy for specific upper-limb configurations.
• While generally reliable, users must be aware of pose and sensor placement limitations to ensure accurate risk assessments.