Author(s): KEISUKE, O., KIYOSHI, H., NAOTO, M., AKIKO, K., WAKO, K., HIROSHI, N., SHUNYA, U., MASAKI, T., Institution: GRADUATE SCHOOL OF DOSHISHA UNIVERSITY, Country: JAPAN, Abstract-ID: 586

INTRODUCTION:
Analyzing the kinematic characteristics of running performance, such as initial contact (IC) and toe-off (TO) timing, step length (SL), and frequency (SF), is a common practice among scientists and practitioners [1, 2]. Outdoor running is commonly measured by setting markers at intervals (e.g., every 50 m) and calculating section averages of these characteristics using a video camera [3]. This study aims to refine the analysis of running characteristics during 400 m runs, focusing on capturing the changes in running characteristics with each step using sensor fusion of GNSS and IMU.
METHODS:
The subject was a male runner specializing in 400 m running. GNSS (Sampling frequency: 100 Hz) and an IMU (1,000 Hz) were attached to the head. The subject completed a 400 m run at maximum effort. Running velocity was measured with differential and real time kinematic (RTK) modes of GNSS. Initial contact and TO timing were detected using IMU acceleration (Y and Z axis) in a global coordinate system estimated by the extended Kalman filter, filtering out centrifugal and tangential accelerations. This process allowed us to identify the exact timestamps for IC and TO. Step length was calculated by synchronizing these timestamps with GNSS (latitude and longitude), defined as the 2D distance between successive IC points. However, the subject’s head tilt at IC, which varied between left and right ICs, affected head position and SL accuracy. To address this, head tilt data from the IMU was incorporated, adjusting for discrepancies due to head movements, ensuring a more precise determination of SL. We employed video analysis to validate SL, IC, and TO timings. The imaging of the subject running was recorded by two cameras (PYTHON 1300) at the curved and straight segments of the track, each over 8 m, with footage analyzed using the software WINanalyze for detailed 3D motion insights. Moreover, the subject running was recorded with another Go pro camera to count the number of steps over 400 m.
RESULTS:
The GNSS data allowed us to monitor 2D velocity changes at 0.01-second intervals over 400 m. Average root mean square error (RMSE) between 3D analysis and RTK GNSS velocities were 0.31 ± 0.29 m/s on the straight segment and 0.19 ± 0.17 m/s on the curved segment. Initial contact and TO timings were measured by IMU with the accuracy within 0.01 seconds (minimal resolution capability of camera measurement) when compared with camera data. The number of steps over the 400 m measured by IMU was identified as 200, which was the same steps measured by Go pro camera. The difference in SL were 0.082 ± 0.037 m in differential GNSS, 0.032 ± 0.016 m in RTK GNSS, and 0.016 ± 0.011 m in the sensor fusion of RTK GNSS and IMU after correcting for head tilt.
CONCLUSION:
It was found that the sensor fusion of GNSS and IMU enables precise and accurate measurements of velocity, IC, and TO timing in each running step with minor errors.

1. Hunter et al. (2004) 2. van Oeveren et al. (2017) 3. Hanon & Gajer (2009)