INTRODUCTION: An optimal trajectory is one of the key factors in achieving higher performance in different skiing disciplines. It is important to analyze such curves to determine the states that correlate to the final result. To obtain such information, the current practice is to put GNSS trackers on the body or skis and use a digital terrain model to map the athlete`s position and data measurement. Computer Vision algorithms applied to videos capturing the performance is an alternative that does not need wearable sensors 1. To the best of our knowledge, no study is present to reconstruct the trajectory of skiers in videos automatically. We present an AI algorithm that computes such evidence on videos acquired from a single unconstrained camera. METHODS: Our algorithm considers a video as a sequence of frames and it worksonline, i.e., it is inputted the latest frame and outputs the trajectory executed by the athlete with the correct perspective with respect to the scene. The algorithm first runs a detector and a visual tracker to automatically find and follow the target skier. Then, an automatic key-point detection and matching algorithm is employed to estimate the camera motion across consecutive frames. This is used to obtain the homography transformation, which maps the points computed by the tracker to the right perspective. RESULTS: Qualitative experiments on broadcasting and smartphone videos featuring alpine skiers and ski jumpers show the algorithm provides immediate visual feedback about the space traversed by the athlete (Figure 1). Our algorithm allows analyses of synchronized trajectories or measurements directly inside the video frames (Figure 2). DISCUSSION/CONCLUSION: The qualitative results achieved show the potential of our solution for quick visual feedback (relevant for broadcasting applications) and more in-depth analyses (for training purposes). Future work will focus on quantifying
© Copyright 2023 9th International Congress on Science and Skiing, March 18 - 22, 2023, Saalbach-Hinterglemm, Austria. Veröffentlicht von University of Salzburg. Alle Rechte vorbehalten.