Body-Worn IMU Human Skeletal Pose Estimation Using a Factor Graph-Based Optimization Framework.

Area of Science:

• Biomechanics
• Human Motion Analysis
• Wearable Sensor Technology

Background:

• Inertial measurement units (IMUs) for human joint angle estimation typically require sensor calibration or specific motions.
• Magnetometers, often used for heading reference in IMU systems, are unreliable indoors, leading to estimation issues.
• Existing IMU methods without magnetometers face unobservability challenges.

Purpose of the Study:

• To propose a novel magnetometer-free human joint angle estimation method for lower body kinematics.
• To address unobservability issues in IMU-based motion tracking without magnetic sensors.
• To enhance the accuracy and robustness of IMU-based human motion analysis.

Main Methods:

• Developed a magnetometer-free IMU estimation approach utilizing lower body kinematics for sufficient degrees of freedom excitation.
• Introduced an expanded lower body model including a novel knee hinge model, anthropometry, and segment length discrepancies.
• Formulated the maximum a posteriori problem as a factor graph and employed post-hoc, on-manifold global optimization for inference.

Main Results:

• Achieved root mean square error (RMSE) of 4.34° for knee flexion/extension angle estimation.
• Demonstrated accuracy comparable to state-of-the-art methods that utilize magnetometers.
• Validated the method on 12 participants performing a prescribed human motion profile task.

Conclusions:

• The proposed magnetometer-free method provides robust lower body joint angle estimation.
• The framework offers desirable observability qualities without relying on potentially unreliable indoor magnetic field data.
• The developed system is extensible for modeling additional joints and constraints in human motion analysis.