Conversion of Upper-Limb Inertial Measurement Unit Data to Joint Angles: A Systematic Review

Zhou Fang, Sarah Woodford, Damith Senanayake, David Ackland2023 · DOI

TL;DR

This systematic review evaluates how Inertial Measurement Units (IMUs) can be used to accurately measure upper-limb joint angles, a crucial task for rehabilitation, sports science, and ergonomics. The study found that while IMUs show high accuracy for elbow and humerothoracic joints, challenges remain with scapulothoracic and glenohumeral joints due to skin motion artifacts.

What Problem Does This Paper Address?

Accurately measuring human motion, particularly the complex movements of the upper limb, is vital across many fields. Traditional methods like optoelectronic motion capture are accurate but expensive, require dedicated spaces, and are time-consuming. Inertial Measurement Units (IMUs) offer a portable, wireless, and more accessible alternative for motion tracking, both in and out of the lab. However, converting raw IMU data (from accelerometers, gyroscopes, and magnetometers) into precise 3D joint angles for the upper limb, especially for joints like the scapula and shoulder, remains a significant challenge. This paper aims to systematically review and quantify the accuracy of different IMU-based methods for upper-limb joint angle estimation compared to the gold standard of optoelectronic systems.

Methods

This paper is a systematic review, meaning it doesn't introduce new experimental methods but rather analyzes existing research. The authors searched five major databases for studies that used IMUs to measure upper-limb joint angles and compared these measurements to optoelectronic motion analysis. They extracted data on:

  • Sensor-to-segment calibration: How IMUs were attached and aligned with body segments.
  • Sensor-fusion algorithms: Techniques (like Kalman filters, complementary filters, and gradient descent) used to combine data from different IMU sensors and reduce noise and drift.
  • Joint-angle calculation methods: How the fused sensor data was translated into joint angles (e.g., Euler angle decomposition, vector geometry).
  • Tasks performed: The types of movements studied.
  • Kinematic errors: Quantitative measures of the difference between IMU-derived and optoelectronic-derived joint angles.

They also assessed the quality of the included studies using a customized checklist.

Datasets

This paper is a systematic review and does not present its own experimental dataset. Instead, it analyzes data from 52 previously published studies that used various datasets for their experiments, typically involving human subjects performing specific upper-limb movements.

Key Insights and Ideas

  • Joint-specific accuracy: The accuracy of IMU-based joint angle measurement varies significantly by joint. Elbow and humerothoracic joints are more accurately measured than the scapulothoracic and glenohumeral joints.
  • Skin motion artifacts are a major hurdle: The movement of skin over underlying bone (especially for the scapula) is a primary source of error for IMU-based measurements of shoulder and scapular joints.
  • Kalman filters and Euler angles are workhorses: Sensor-fusion algorithms, particularly Kalman filters, and Euler angle decomposition are frequently used and effective for humerothoracic and elbow joint angle calculations.
  • Functional calibration is key: Methods that involve actual joint movements for calibration (functional joint movements) tend to yield better results than purely static calibrations, especially for more complex joints.
  • Standardization is lacking: There isn't a single, universally adopted approach for IMU-based upper-limb motion analysis, leading to variability in results.

Key Results

  • Overall Accuracy: The review included 52 studies. The methodological quality scores ranged from 9 to 22, with an average of 17.6.
  • Humerothoracic Joint: Achieved relatively high accuracy, with RMS errors around 4° when compared to optoelectronic systems.
  • Elbow Joint: Showed even higher accuracy, with RMS errors as low as 2° for flexion-extension, particularly when using functional joint axis calibration and kinematic constraints.
  • Glenohumeral Joint: Generally showed RMS errors of 6° or higher.
  • Scapulothoracic Joint: Faced the most significant challenges, with RMS errors exceeding 10° for protraction-retraction and anterior-posterior tilt. Some specific tasks, like a side reach, showed very high errors (e.g., 43° for protraction-retraction).

Limitations and Future Work

  • Limitations:
    • Variability in optoelectronic motion analysis protocols (marker placement, coordinate system definitions) makes direct comparison between studies challenging.
  • Inconsistent use of error metrics (RMS error, peak error, mean absolute error) across studies.
  • The use of different IMU models with varying sample rates and sensitivities can affect accuracy.
  • The study primarily focused on activities of daily living, and results might differ for high-speed or impact sports.
  • Future Work:
    • Implement functional joint axis calibrations and IMU-based scapula locators to mitigate skin motion artifacts.
  • Explore artificial neural networks (ANNs) and data-driven approaches for more robust and potentially real-time IMU-to-joint-angle conversion.
  • Develop standardized protocols for IMU-based motion analysis.
  • Further research into machine learning and AI for analyzing large IMU datasets to improve accuracy and generalizability.

Summary updated

BibTeX
@misc{fang2023conversion,
  author = {Zhou Fang and Sarah Woodford and Damith Senanayake and David Ackland},
  title = {{Conversion of Upper-Limb Inertial Measurement Unit Data to Joint Angles: A Systematic Review}},
  year = {2023},
  doi = {10.3390/s23146535},
}

Loading related content...

Figure