INTRODUCTION: While snowboarding was originally only performed in mountainous regions, the popularity of this sport combined with the unpredictability of natural snowfall has led to the creation of dry slopes, i.e., man-made artificial surfaces that simulate the feel of snow, allowing for year-round snowboarding practice and training. To enhance the experience at the dry slope, we've developed an automated video system that provides detailed jump analysis. Using three cameras, we capture athletes from multiple angles and generate zoomed-in videos for review. Our analysis reports, distilled from these videos, includes key metrics like airtime, takeoff speed, height, and distance. METHODS: We automatically clip video segments by predicting the athlete's three dimensional flight path. This prediction is used to project the athlete's estimated position onto the image plane, allowing us to frame the athlete effectively. To achieve this, we track the athlete's movement on the kicker using human pose estimation, which enables us to estimate his takeoff position and velocity. Using the estimated takeoff parameters, we approximate the athlete's flight trajectory using projectile motion with air resistance. To streamline the analysis, we divide the jump into four stages: (1) the inrun, (2) takeoff, (3) airtime, and (4) landing stages, as illustrated in Figure 1. RESULTS/DISCUSSION: Our framing method, which achieved zoom levels of 2x to 4x, provides a solid foundation for future research on snowboarding jump feedback. This includes applications requiring higher resolution footage, such as 3D human pose estimation and scene reconstruction. This could be achieved by utilizing the predicted trajectories to automatically steer a high resolution PTZ-camera (at which point timing and efficient computation becomes more important). Additionally, our analysis results complement subjective visual observations with objective metrics. However, further testing is needed to verify their accuracy. CONCLUSION: This research provides a foundation for 3D human pose estimation of snowboarding jumps by using multi-camera video footage to estimate the athlete's flight trajectory. Future work can build upon this foundation by implementing automatic athlete tracking, enabling higher zoom levels and improved image resolution. Furthermore, a detailed jump analysis, including airtime, takeoff velocity, height, and distance, can be derived from the predicted trajectory. While this methodology was only demonstrated at a dry slope in Genk (Belgium), it can serve as a model/basis for similar research elsewhere.
© Copyright 2025 10th International Congress on Science and Skiing, January 28 - February 1, 2025, Val di Fiemme, Italy. Alle Rechte vorbehalten.