Internal Impact Validation: Technical Report

Several wearable athlete tracking systems provide a collision variable calculated from unvalidated algorithms. In lieu of using an automatic collision detection algorithm, we sought to quantify impact frequencies directly from raw accelerometer data measured by one of these systems (EVO, Catapult Sports, Australia). The aim of this technical report is to provide a framework for practitioners to follow in the detection of contact-based events in sport.

A smaller subsample of data was used in the internal validation process. A random sample of 15 player-game files were used in this analysis and represent examples from each position, multiple games and across both seasons. All procedures were approved by the La Trobe University Human Research Ethics Committee (HEC19375).

The internal validation process occurred over five steps (Data combination, threshold detection, threshold - secondary confirmation, filtering, filtering - secondary confirmation). Following the combination of positioning (GPS) and accelerometer datafiles, a process was applied to find the optimal impact detection threshold and then smoothing filter. This process allows for detected impacts to correspond to contact-based events, validated by matching data with OPTA video-coded files, and removes instances of impacts due to other movements such as a heavy foot strike or change of direction.

In conclusion the internal validation process found that a g-force value of 5 g and a 0.1 s smoothing filter was accurate at detecting impacts in professional rugby union players. When matched with video coded footage, this resulted in a FPR of 9% and a FNR of 13%.