Handball Action Dataset

Details:

• Recording Setup: Videos were shot from different angles with GoPro cameras mounted on the playground’s left or right side.
• Player Details: On average, each scene features 12 players, with one or two players performing the action of interest.
• Action Extraction: Actions were isolated using Mask RCNN detector and Deep SORT for tracking, followed by manual selection of the appropriate tracked player.

Dataset Structure:

• scenes/: Contains mp4 videos with actions performed in the context of other players. Subdirectories are organized by action class.
• actions/: Contains videos of isolated actions by single players, organized by performed action class. File names correspond to the original video from which the action was extracted.

Additional Content: This dataset is invaluable for training AI models in sports analytics, player behavior analysis, and performance enhancement. By providing isolated action clips, it enables precise action recognition and analysis, facilitating the development of robust computer vision models tailored for sports applications.

Conclusion

This dataset provides a strong foundation for developing machine learning models for action recognition in sports videos. It supports applications in sports analytics, player performance evaluation, and activity tracking. With its structured video data and diverse actions, it enables accurate and reliable computer vision solutions.

This dataset is sourced from Kaggle.