Abstract
The purpose of this research was to develop a wearable, low-cost prototype based on real-time kinematic GPS and a microelectromechanical inertial measurement unit to measure the sprinting velocity of an athlete. The software package RTKLIB was used to calculate the RTK-GPS positions and different Kalman filters were implemented to provide a loosely coupled sensor fusion. With this setup, we performed empirical studies to determine whether the velocities obtained by this novel approach are sufficiently accurate for a performance orientated training. Therefore, field tests for 30- to 400-m sprint distance were conducted with simultaneous measurements with different reference systems, such as a laser device or timing gates. The evaluation revealed a correspondence between prototype and reference systems with distance and timing errors of \(\pm \, 2\,\%\) and high correlations for the velocities (R = 0.996, P <0.001) for 68 % of the trials. However, for remaining 32 % of the trials no acceptable performance parameters could be obtained due to GPS problems. Overall, the developed prototype showed great potential and might allow closing the gap between the accuracy and flexibility of the established reference systems as soon as its susceptibility to GPS problems is lowered.
Similar content being viewed by others
References
Akenhead R, French D, Thompson KG, Hayes PR (2013) The acceleration dependent validity and reliability of 10 Hz GPS. J Sci Med Sport
Bezodis N, Trewartha G (2012) Measurement error in estimates of sprint velocity from a laser displacement measurement device. Int J Sports Med 33:439–444
Bhatt D, Aggarwal P, Devabhaktuni V, Bhattacharya P (2014) A novel hybrid fusion algorithm to bridge the period of GPS outages using low-cost INS. Expert Syst Appl 41(5)
Brodie M, Walmsley A, Page W (2008) Fusion motion capture: a prototype system using inertial measurement units and GPS for the biomechanical analysis of ski racing. Sports Technol 1:17–28
Caron F, Duflos E, Pomorski D, Vanheeghe P (2006) GPS/IMU data fusion using multisensor Kalman filtering: introduction of contextual aspects. Inf Fusion 7:221–230
Castellano J, Casamichana D, Calleja-Gonzlez J, Ostojic SM (2011) Reliability and accuracy of 10 Hz GPS devices for short-distance exercise. J Sports Sci Med
Chelly SM, Denis C (2001) Leg power and hopping stiffness: relationship with sprint running performance. Med Sci Sports Exerc 33(2):326–33
Coutts AJ, Duffield R (2010) Validity and reliability of GPS devices for measuring movement demands of team sports. J Sci Med Sport 13(1):133–135
Elisson V, Gaessler G (2014) Low cost relative GNSS positioning with IMU integration. Master’s thesis. Chalmers University of Technology
Faragher R (2012) Understanding the basis of the Kalman Filter via a simple and intuitive derivation. In: IEEE Signal Processing Magazine, pp 128–132
Fusionsport. Timing accuracy: are you up to speed? 2010. http://www.fusionsport.com/support/product-information/research-tech-specs?download=54:timing-accuracy-research-update
Haugen T, Buchheit M (2016) Sprint running performance monitoring: methodological and practical considerations. Sports Med 41:433–521
Hay JG (1993) The biomechanics of sports techniques, 4th edn. Prentice Hall, New Jersey
Hosseini S-A, Farrokhi M (2009) New steady state Kalman Filter for tracking high maneuvering targets. http://www.iust.ac.ir/files/ee/pages/maghalat/5.pdf
Hunter JP, Marshall RN, McNair PJ (2004) Interaction of step length and step rate during sprint running. Med Sci Sports Exerc 36(2):261–271
Johnston RJ, Watsford ML, Kelly SJ, Pine MJ, Spurrs RW (2012) The validity and reliability of 10 Hz and 15 Hz GPS units for assessing athlete movement demands. J Strength Cond Res
Nagahara R, Botter A, Rejc E, Koido M, Shimizu T, Samozino P, Morin J-B (2016) Concurrent validity of GPS for deriving mechanical properties of sprint acceleration. Int J Sports Physiol Perform 12(1):129–132
Rabbou MA, El-Rabbany A (2014) Tightly coupled integration of GPS-PPP and MEMS-based inertial system using EKF and UKF. FIG Congress 2014
Takasu T (2016) rtklib. http://www.rtklib.com/rtklib.htm
Tan H, Wilson AM, Lowe J (2008) Measurement of stride parameters using a wearable GPS and inertial measurement unit. J Biomech 41:1398–1406
Tawk Y, Tom P, Botteron C, Stebler Y, Farine P-A (2014) Implementation and performance of a GPS/INS tightly coupled assisted PLL architecture using MEMS inertial sensors. Sensors 14:3768–3796
Varley M, Fairweather I, Aughey R (2012) Validity and reliability of GPS for measuring instantaneous velocity during acceleration, deceleration and constant motion. J Sports Sci 30:121–127
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
This appendix provides the mathematical descriptions of the recursive Kalman filters KF2 and KF5, which are based on the alternating execution of the prediction (Eq. 1) and correction steps (Eq. 2).
The mainly used movement model is the constant-jerk model (CJ) with the state vector \(\hat{x}\) (Eq. 3) and the dynamic matrix A (Eq. 4).
Removing the jerk tate leads to the constant-acceleration model (CA). The uncertainty matrix P is initialised with the identity matrix and the measurement vector H is set to \(H = (1~0~1~0)\).
1.1 KF2
For KF2, the CJ model is used, taking the horizontal displacement and the IMU’s forward acceleration as inputs. The measurement covariance matrix R (Eq. 5) depends on the current GPS-error \(\sigma GPS\) (provided by the GPS receiver) and the IMU’s standard deviation. As the upper body rotation causes additional noise, it is increased from 0.01 \(\frac{m}{s^2}\) to 0.1 \(\frac{m}{s^2}\). The process–noise–variance matrix Q (Eq. 6) was determined experimentally below.
1.2 KF5
KF5 is a combination of six Kalman filters. The first one utilizes a CA model with state vector \(\hat{x}_Y\) (Eq. 7) to maintain a drift-free yaw estimation, using the yaw angle provided by the GPS and the yaw-angle velocity provided by the IMU as inputs. The measurement noise for the GPS heading and the process–covariance matrix \(Q_Y\) (Eq. 9) had to be determined experimentally. However, the gyroscope has a stated variance of \(0.1^\circ\)/s with a bias of ± 2.5 \(^\circ\). A selected measurement variance of \(5^\circ\)/s in \(R_Y\) (Eq. 8) is, therefore, appropriate for around 2 h of measuring.
The second Kalman filter utilizes again a CA model with state vector \(\hat{x}_B\) (Eq. 11) to smoothen the current yaw bias of the IMU. The only input is the bias itself (Eq. 10), and \(R_B\) (Eq. 12) and \(Q_B\) (Eq. 13) were determined experimentally.
In a next step, the bias \(\hat{x}_{B_1}\) is used in a quaternion rotation to nullify the IMU’s yaw drift and align the accelerations provided by the IMU with the north–east–up axis. With three further Kalman filters utilizing the CJ model, the GPS northing, easting and height are then separately fused with the corresponding acceleration measurement of the IMU. Therefore, the same process–noise–variance matrix Q is used as in KF2 (Eq. 6), the measurement covariance matrix R (Eq. 5) is adapted to the GPS-error on the specific axis and the IMU variance is lowered to the original value of 0.01 \(\frac{m}{s^2}\). These three Kalman filters provide the displacement and velocity in every three directions, but not on the desired horizontal plane. Thus, one last Kalman filter with a CA model and the velocity as only input state is required in KF5. The covariance matrix \(R_V\) (Eq. 14) and sigma \(\sigma _{qV} = 0.05\) for \(Q_V\) (Eq. 15) had to be determined experimentally and the horizontal velocity, as input parameter, is the combined velocity towards north and east (Eq. 16).
Rights and permissions
About this article
Cite this article
Mertens, J.C., Boschmann, A., Schmidt, M. et al. Sprint diagnostic with GPS and inertial sensor fusion. Sports Eng 21, 441–451 (2018). https://doi.org/10.1007/s12283-018-0291-0
Published:
Issue Date:
DOI: https://doi.org/10.1007/s12283-018-0291-0