INTRODUCTION: In slalom skiing, as in any other alpine skiing competition, the fastest time decides the winner. Athletes as well as trainers are therefore always interested in a detailed quantification of the performance. Are there certain passages where time was lost or gained? An informative measurement is the gate-to-gate time which states the time passed between slalom gates. Ideally, these measurements are available to the athlete and trainer right after the run for immediate feedback. Different strategies to measure gate-to-gate time exist. However, they are either infeasible with respect to the amount of work needed, e.g., in manual video analysis; or they lack accuracy, e.g., taking the total time divided by number of gates (Swarén, M. et al., 2021); or rules prohibit the use of the required sensors, e.g., magnetic sensors in gloves or gates (Fasel, B. et al., 2019). In this work we present a new concept to measure gate-to-gate time in fully automated fashion from audio recordings. METHODS: When the athletes passe a gate, they hit it which creates two distinctive sounds: one at contact with the gate as well as one whipping noise when the gate hits the snow. We train a convolutional neural network on a data set of audio recordings of the slalom race in Garmisch-Partenkirchen 2022 to detect the sound pattern in the audio automatically. This training data is labeled using video recordings by tagging the gate contact frame by frame. RESULTS: The trained network detects and computes the gate-to-gate timing for new unlabeled runs with high accuracy. We validate the accuracy of the results as well as generalizability by comparing them to labeled validation and test data. DISCUSSION/CONCLUSION: The network provides timing to athlete and trainer immediately after the run. This instantaneous feedback helps to improve performance and to identify tactical choices. We also discuss challenges in the approach.
© Copyright 2023 9th International Congress on Science and Skiing, March 18 - 22, 2023, Saalbach-Hinterglemm, Austria. Veröffentlicht von University of Salzburg. Alle Rechte vorbehalten.