The Design of GNSS/IMU Loosely-Coupled Integration Filter for Wearable EPTS of Football Players

Mingu Kim, Chulwoo Park, Jinsung Yoon2023 · DOI

TL;DR

This paper presents a novel loosely-coupled integration filter for wearable Electronic Performance and Tracking Systems (EPTS) in football, combining Global Navigation Satellite System (GNSS) and Inertial Measurement Unit (IMU) data. The proposed filter offers smoother and more accurate tracking of player movements compared to GNSS alone, crucial for performance analysis and injury prevention in dynamic sports.

What Problem Does This Paper Address?

In modern sports science, collecting and analyzing athlete performance data is paramount for improving player performance and reducing injury risks. Wearable Electronic Performance and Tracking Systems (EPTS) are increasingly used for this purpose, often integrating GNSS and IMU sensors. While GNSS provides global positioning, its low data rate and susceptibility to signal loss limit its effectiveness for capturing the rapid, dynamic movements of athletes. IMUs, on the other hand, offer high-frequency data but suffer from accumulating drift errors over time. This paper addresses the challenge of effectively fusing these two sensor types to achieve accurate, high-frequency, and robust tracking of football players' movements, overcoming the limitations of each sensor individually.

Methods

The core of this work is the design of a loosely-coupled integration filter for GNSS and IMU data. This approach is chosen because commercial wearable EPTS typically provide processed GNSS position and velocity data, rather than raw measurements, making a loosely-coupled strategy more practical. The filter utilizes an Extended Kalman Filter (EKF) to handle the nonlinear dynamics of integrating GNSS and IMU measurements. A key innovation is the use of quaternions for attitude estimation, which effectively avoids the gimbal lock singularity that can occur with Euler angles. The filter's state vector includes position, velocity, and attitude (represented by quaternions). The prediction step uses IMU data (accelerations and angular rates) to propagate the state, while the update step fuses GNSS position and velocity measurements to correct the predicted state. Attitude estimation is performed separately using a Kalman filter on accelerometer and gyroscope data.

Datasets

The paper does not describe datasets used for training. Instead, it focuses on experimental validation using real-world data. The experiments involved:

  • Rover Field Tests: An autonomous rover equipped with the wearable EPTS was used to test the filter's performance in controlled driving scenarios (circular and straight paths).
  • Athlete Field Tests: Football players wearing the EPTS performed various movements, including sprinting and running back and forth.
  • Comparison with Commercial Systems: The performance of the proposed filter was compared against a commercial high-precision GNSS/IMU AHRS (Microstrain's 3DM-GX5).
  • Ground Truth Validation: A Vicon motion capture system was used as the ground truth for validating the accuracy of the designed integration filter, particularly for distance and speed measurements during sprinting.

Key Insights and Ideas

  • Loosely-coupled is practical for wearables: The paper demonstrates that a loosely-coupled approach is a viable and effective strategy for integrating GNSS and IMU data in wearable EPTS, especially when raw GNSS measurements are not available.
  • Quaternion for robust attitude estimation: Using quaternions successfully circumvents the gimbal lock issue, providing a more stable and reliable attitude estimation compared to Euler angles.
  • Smoother trajectories and higher data rates: The integration filter significantly smooths GNSS trajectories and provides data at a much higher rate (100 Hz from IMU vs. 10 Hz from GNSS), enabling more detailed analysis of instantaneous movements.
  • Impact of athlete dynamics: The study highlights how dynamic athlete movements, like sprinting and foot strikes, can introduce noise and errors into IMU measurements, affecting the accuracy of the integrated system. This points to the need for robust filtering techniques that can handle these transient effects.

Key Results

  • Smoother Trajectories: The integration filter produced significantly smoother trajectories compared to raw GNSS data in both rover and athlete tests.
  • Improved Speed Estimation: The estimated speed from the integration filter showed much smaller deviations compared to GNSS speed data. For the rover, the speed was estimated at 1 m/s with minimal deviation. For athletes, the speed increased up to 8 m/s.
  • Accurate Distance Tracking: Vicon validation showed that the designed integration filter achieved distance errors of less than or equal to 10 cm (0.1 m) across various sprinting distance zones, indicating reliable distance estimation.
  • Comparable Performance to High-End Systems: The proposed filter's performance was comparable to a commercial high-precision GNSS/IMU AHRS in terms of trajectory and velocity estimation.
  • Attitude Estimation Similarity: Estimated attitudes (roll and pitch) from the proposed filter were similar to those from the commercial AHRS, though yaw angle estimation showed some differences, potentially due to the limited use of gyroscope data for yaw in the EPTS.

Limitations and Future Work

  • Yaw Angle Estimation: The paper acknowledges that the yaw angle estimation might have differences compared to high-precision systems. They suggest that future research could explore the use of magnetometers to improve yaw angle estimation.
  • Impact of Athlete Dynamics: The study noted that athlete-specific dynamics (e.g., foot strikes during sprinting) can affect IMU measurements and introduce errors. Further research could focus on more advanced filtering techniques to mitigate these effects.
  • Advanced Performance Data: The authors suggest that future work could focus on extracting more advanced performance data, such as the number of steps and stride velocity, from the integrated GNSS/IMU data.
  • RTK-GNSS Integration: Integrating Real-Time Kinematics (RTK)-GNSS with IMU is proposed as a future direction to potentially enhance position accuracy.

Summary updated

BibTeX
@misc{kim2023design,
  author = {Mingu Kim and Chulwoo Park and Jinsung Yoon},
  title = {{The Design of GNSS/IMU Loosely-Coupled Integration Filter for Wearable EPTS of Football Players}},
  year = {2023},
  doi = {10.3390/s23041749},
}
Figure