Abstract

Skiing is a popular winter sport discipline with a long history of competitive events. In this domain, computer vision has the potential to enhance the understanding of athletes’ performance, but its application lags behind other sports due to limited studies and datasets. Our paper makes a step forward in filling such gaps. A thorough investigation is performed on the task of skier tracking in a video capturing his/her complete performance. Obtaining continuous and accurate skier localization is preemptive for further higher-level performance analyses. To enable our study, the largest and most annotated dataset for computer vision in skiing, SkiTB, is introduced. Several visual object tracking algorithms, including both established methodologies and a newly introduced skier-optimized baseline algorithm, are tested using the dataset. The results provide valuable insights into the applicability of different tracking methods for vision-based skiing analysis.

WACV Paper

CVIU Paper

Dataset Stats

300
Multi-Camera Videos
2019
Single-Camera Videos
352978
Frames
196
Athletes
161
Locations

Videos and Annotations

Qualitative examples of some of the video contained and annotated in SkiTB. The videos cover the skiers’ complete performance. Each video is densely labeled with the bounding-boxes of a single target skier, and with attributes identifying the camera ID, the visual changes that the skier undergoes, the type of skiing discipline, the athlete ID, the location of the competition, the weather conditions, as well as the parameters of the skiing performance. Different training and test splits are also provided.

Results

Overall and per-skiing discipline results in the multi-camera (MC) tracking setting. The F-Score ↑, Precision ↑, and Recall ↑ scores are presented for each of the studied tracking algorithms. In general, we observe that ski jumping (JP) is the discipline in which trackers perform better, followed by alpine skiing (AL). Freestyle skiing (FS) offers the most challenging situations.
Qualitative examples of SkiTB's tracking sequences along with the results of the top-performing trackers.

Downloads

References

 
@InProceedings{SkiTBwacv,
  author = {Dunnhofer, Matteo and Sordi, Luca and Martinel, Niki and Micheloni, Christian},
  title = {Tracking Skiers from the Top to the Bottom},
  booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  month = {Jan},
  year = {2024}
}

@article{SkiTBcviu,
  title = {Visual tracking in camera-switching outdoor sport videos: Benchmark and baselines for skiing},
  journal = {Computer Vision and Image Understanding},
  volume = {243},
  pages = {103978},
  year = {2024},
  doi = {https://doi.org/10.1016/j.cviu.2024.103978},
}

Acknowledgements

This research has been supported by the project between the University of Udine and the organizing committee of EYOF 2023 Friuli-Venezia Giulia.

Copyright © Machine Learning and Perception Lab - University of Udine - 2023 - 2024

Website template was inspired by TAO Dataset.