Development of an IMU-based motion capture system for swimming: A study protocol

Abstract

Optical motion capture systems are the gold standard for collecting three-dimensional kinematic data in sports. However, while their use in flow channels is somewhat practical, they are difficult to employ in swimming pools, e.g. due to camera placement and the dynamic refraction of light and air bubbles. IMUs can circumvent these problems, but have so far primarily been used to calculate relatively condensed information such as distance per stroke or lap counts (Magalhaes et al., 2015). Additionally, current commercially available IMU-based systems for the collection of segment orientations and joint angles are often bulky and not waterproof (Ascenso, 2021). Therefore, the purpose of this study was to develop a novel IMU-based motion capture system for the specific requirements of swimming.

Hardware
State-of-the-art IMUs (ICM-20948, InvenSense Inc., San José, California, USA) in combination with microcontrollers (ESP32-S3, Espressif Systems, Shanghai, China) are used to record raw IMU signals. Each unit (i.e. IMU and microcontroller) is sealed in foil using a vacuum sealer to waterproof the electronics and remove air. The microcontrollers store the IMU raw data internally (up to 3.8 MB) during a recording and allow for a later data transmission to external systems. To do this, each microcontroller creates its own asynchronous web server, which is used to send commands to the microcontrollers and retrieve data via HTTP requests.

Data processing
An accurate IMU calibration is particularly important, as IMUs, such as the ICM-20948, contain systematic errors and are susceptible to noise. The calibration process used in this study is largely based on Särkkä et al. (2017). The raw sensor outputs are compared with known inputs (reference signals): net rotations between known positions, gravity and the magnetic field of the Earth. For this, the sensors are rotated into 24 different positions. Successive positions differ by 90° and the sensors remain stationary for a few seconds in each position. Using an iterative algorithm, 12 calibration parameters are calculated for each accelerometer and magnetometer and three for the gyroscope. After calibration, the data is processed using a 5 Hz low-pass Butterworth filter and the NXP fusion filter (NXP Semiconductors, 2016) to determine the orientations of the sensors relative to a global coordinate system. Sensor-to-segment calibration is achieved by averaging the rotations during a neutral standing position. The microcontrollers are synchronized by repeatedly exchanging timestamps and calculating linear regressions with RANSAC (i.e. ignoring outliers) that relate the local timestamps to a common global time.

Validation and testing
We present validation results for 3D shoulder, elbow and head angles for simulated front crawl swimming as well as isolated movements in the five joints using a synchronized optical motion capture system in the laboratory. In addition, findings from plausibility tests during actual swimming in a swimming pool are presented.

References

Ascenso, G. (2021). Development of a non-invasive motion capture system for swimming biomechanics (Doctoral dissertation). Manchester Metropolitan University, United Kingdom.

NXP Semiconductors. (2016). AN5023 Sensor fusion kalman filters. https://www.nxp.com/docs/en/application-note/AN5023.pdf

Magalhaes, F. A. de, Vannozzi, G., Gatta, G., & Fantozzi, S. (2015). Wearable inertial sensors in swimming motion analysis: A systematic review. Journal of Sports Sciences, 33(7), 732–745. https://doi.org/10.1080/02640414.2014.962574

Särkkä, O., Nieminen, T., Suuriniemi, S., & Kettunen, L. (2017). A multi-position calibration method for consumer-grade accelerometers, gyroscopes, and magnetometers to field conditions. IEEE Sensors Journal, 17(11), 3470–3481. https://doi.org/10.1109/JSEN.2017.2694488