Influence of Visual-Inertial Sensor-to-Segment Calibration on Upper Limb Joint Angles Estimation

Artificial Intelligence Information Science Automation Orthopedics Sports Medicine Medicine
visual-inertial sensors joint angles estimation calibration methods multi-body inverse kinematics upper limb rehabilitation

Upper Limb Joint Angle Estimation Based on Visual-Inertial Sensors

Research on Upper Limb Joint Angle Estimation Based on Visual-Inertial Sensors and the Impact of Calibration Methods

Academic Background

Upper limb dysfunction, especially in post-stroke patients, significantly impacts their ability to perform daily activities. Rehabilitation training is a critical method for restoring upper limb function, but its effectiveness often relies on the accurate assessment of joint angles. Traditional optical motion capture (OMC) systems are the “gold standard” for joint angle estimation, but their high cost and bulkiness limit their widespread use in real clinical settings. In recent years, low-cost sensors such as visual-inertial measurement units (VIMUs) have emerged as a promising alternative. However, their inherent measurement inaccuracies and calibration issues hinder their clinical applications.

This study aims to explore the impact of different calibration procedures, inverse kinematics (IK) methods, and measurement modalities on the accuracy of upper limb joint angle estimation during rehabilitation training. The core question is: How can reasonable calibration procedures and IK methods reduce the measurement errors of VIMUs, thereby making them practical for clinical rehabilitation?

Source of the Paper

This paper was co-authored by Mohamed Adjel, Raphael Dumas, Samer Mohammed, and Vincent Bonnet from research institutions such as NaturalPad, LISSI, the University of Lyon, and LAAS-CNRS in France. The research was published in IEEE Transactions on Automation Science and Engineering and has been accepted for publication in 2025.

Research Design

Overall Framework of the Study

The study primarily consists of the following steps:

  1. Sensor-to-Segment Calibration:

    • Anatomical Calibration: Establishing local coordinate systems for limb segments using fixed anatomical landmarks.
    • Functional Calibration: Estimating joint centers and longitudinal axes of limb segments through specific joint movements.

  2. Construction of the Upper Limb Biomechanical Model:

    • A biomechanical model of the upper limb with 7 degrees of freedom was established, following the joint coordinate system recommendations of the International Society of Biomechanics (ISB).

  3. Comparison of Inverse Kinematics (IK) Methods:

    • Multi-body IK: Based on the biomechanical model, it simultaneously estimates the poses of all limb segments, considering joint constraints.
    • Single-body IK: Estimates the rotation of each limb segment separately, ignoring joint constraints.

  4. Data Collection and Experimental Setup:

    • Seven healthy young volunteers participated in three rehabilitation tasks (Pick and Place, Ruler, Bottle Task).
    • Data were synchronously collected using VIMUs and OMC systems for comparative analysis.

  5. Results Analysis and Comparison:

    • The accuracy of joint angle estimation obtained using different calibration procedures and IK methods was compared, with key metrics being the root mean square error (RMSE) and Pearson correlation coefficient ®.

Research Results

  1. Impact of Calibration Procedures on Joint Angle Estimation:

    • When the same calibration procedure was used, the joint angle estimation errors between VIMU and OMC data were small (RMSE ≤ 7.9 degrees, r ≥ 0.86).
    • When different calibration procedures were used, the errors increased significantly (RMSE > 10 degrees), highlighting the importance of calibration consistency for accuracy.

  2. Comparison of Inverse Kinematics Methods:

    • Multi-body IK demonstrated higher robustness when processing VIMU data, with significantly lower errors than single-body IK.
    • The RMSE of multi-body IK was as low as 2.7 degrees, while that of single-body IK reached 5.5 degrees.

  3. Impact of Calibration Offsets:

    • Removing calibration offsets significantly reduced the RMSE of joint angle estimation, indicating that calibration offsets are a major source of error.

  4. Comparison Between Anatomical and Functional Calibration:

    • Functional calibration had slightly lower errors than anatomical calibration, suggesting that functional calibration offers certain advantages in robustness.

Research Conclusions

This study demonstrates that the consistency of calibration procedures is crucial for the accuracy of joint angle estimation using VIMUs, especially in clinical rehabilitation. The use of multi-body IK methods can effectively reduce measurement errors. As a low-cost sensor, VIMUs, when properly calibrated and applied with multi-body IK, can provide joint angle estimation accuracy comparable to OMC systems. This opens up possibilities for their use in real clinical settings.

Research Highlights

  1. Importance of Calibration Procedures: The study emphasizes the importance of maintaining consistency during calibration, particularly when using low-cost sensors.
  2. Advantages of Multi-body IK: Multi-body IK demonstrated higher robustness and accuracy when processing VIMU data, offering a more reliable technical approach for clinical rehabilitation.
  3. Potential of Low-Cost Sensors: By optimizing calibration procedures and IK methods, low-cost sensors such as VIMUs can play a significant role in rehabilitation, reducing healthcare costs.

Research Value

This study not only provides a low-cost sensor-based method for joint angle estimation in clinical rehabilitation but also offers important insights for future research. Future studies could further validate the applicability of this method in more complex and diverse clinical scenarios and explore its potential for whole-body motion analysis.

Additional Information

The main limitations of this study are that the experimental tasks focused on upper limb movements and did not involve whole-body motions. Additionally, the study was conducted under ideal lighting and measurement conditions, and future work is needed to verify the robustness of the method in more challenging environments.

Through this research report, we have provided a detailed analysis of the study’s design, implementation, results, and its potential applications in clinical rehabilitation. This research offers significant technical support and theoretical foundations for the use of low-cost sensors in the field of rehabilitation.