TL;DR
This paper addresses the growing challenge of selecting the right Inertial Measurement Unit (IMU) for pervasive healthcare applications, particularly gait analysis. The authors propose a structured guideline for IMU selection and provide a comparative analysis of seven commercial IMUs, offering practical insights into their specifications, data collection, and raw data quality to aid researchers in making informed choices.
What Problem Does This Paper Address?
Inertial Measurement Units (IMUs) are increasingly vital for pervasive healthcare applications, enabling tasks like localization and movement tracking, with gait analysis being a prominent example. However, the market is flooded with a wide variety of IMU devices, each with different specifications and data collection procedures. This makes it increasingly difficult for researchers to select the most suitable IMU for their specific needs. Existing studies often lack detailed comparisons of IMU features and data quality, making it hard to predict performance before hands-on experience. This paper aims to bridge this gap by providing a clear selection framework and a practical comparison of commonly used IMUs.
Methods
The authors propose a three-step guideline for IMU selection:
- Initial Selection: This involves collecting and confirming IMU specifications (e.g., sensor ranges, sampling rates, memory, battery) and understanding data collection procedures. This step can be completed before acquiring the devices.
- Explore Raw Data: After obtaining the IMUs, this step focuses on gaining hands-on experience with the data collection workflow and assessing the quality of the raw data. This includes checking timestamps, baseline sensor values, and maximum recording times.
- Test on Use Cases: The final step involves testing the IMUs with a specific use-case algorithm. In this study, a gait analysis algorithm was used to evaluate the accuracy of spatio-temporal gait parameters derived from each IMU's data. This allows for a direct comparison of how well each IMU performs in a relevant application.
The gait analysis algorithm used is based on the zero-velocity update technique and an error-state Kalman filter to cope with IMU drift and estimate displacement.
Datasets
The paper does not explicitly mention training datasets. For evaluation, the authors used data collected from seven commercial IMU devices during a gait analysis experiment. The ground truth for comparison was obtained from the OptoGait system, a validated optical gait analysis system.
Key Insights and Ideas
- Structured Selection is Crucial: The paper emphasizes that a systematic approach, moving from specifications to raw data exploration and finally to use-case testing, is essential for effective IMU selection.
- Raw Data Quality Matters: While specifications provide a starting point, the actual quality of raw data (e.g., sampling rate consistency, absence of clipping) significantly impacts the accuracy of derived parameters, especially for complex analyses like spatio-temporal gait estimation.
- Use-Case Specific Testing is Paramount: An IMU that performs well in one aspect (e.g., raw data exploration) might not perform optimally in a specific application like gait analysis. Testing with the intended algorithm is key.
- Calibration is Not Always Enough: Even with calibration options, some IMUs (like MMR) showed performance issues that suggested a need for more precise, user-performed calibration.
- Data Loss is a Major Concern: Streaming IMU data via Bluetooth (as with EXLs3) can lead to significant data loss, disrupting trajectory calculations and impacting accuracy, especially for spatial parameters.
Key Results
- Stride Time Accuracy: Most tested IMUs demonstrated similar high accuracy in estimating stride times, with correlation coefficients generally above 0.94 and limits of agreement below 0.11 seconds.
- Stride Length Variability: Stride length estimation showed more variability. Move 4, Gait Up, and EXLs3 IMUs performed comparably well, with high correlation and agreement.
- Bonsai IMU Acceleration Clipping: The Bonsai IMUs exhibited acceleration data clipping at ±4 g, leading to underestimated stride lengths, particularly for longer strides.
- MMR IMU Clustering: MMR IMUs showed distinct clusters in stride length estimations, suggesting inconsistent performance between the left and right feet, potentially requiring more precise calibration.
- Portabiles IMU Accuracy: Portabiles IMUs showed generally lower accuracy in stride length estimation compared to others, with a larger limit of agreement (0.21 m).
- EXLs3 Data Loss Impact: Data loss from Bluetooth streaming significantly disrupted stride trajectories for the EXLs3 IMU, impacting analysis.
Limitations and Future Work
- Limited Scope of Use Case: The gait analysis algorithm used is just one example. Researchers should consider their specific algorithms and use cases when selecting IMUs.
- Generalizability: The findings are specific to the tested IMUs and the gait analysis use case. Different applications or IMU models might yield different results.
- Raw Data Exploration vs. Performance: The paper notes that initial raw data explorations (sampling intervals, baseline values) did not strongly correlate with the IMUs' performance in the gait analysis test, suggesting that more sophisticated calibration analysis might be needed for a deeper understanding of error characteristics.
- Future Work: The authors suggest that researchers should generalize the conclusions with caution and consider other factors like IMU placement, algorithm choice, and calibration procedures for their specific research topics. They also encourage exploring new IMU models and algorithms as the market evolves.
Summary updated