Skip to content

SmartCrop-H: AI-Based Cropping for Ice Hockey

ice hockey

Overview

SmartCrop-H is an advanced AI-based video cropping solution tailored specifically for the fast-paced world of ice hockey. Now in production with Forzasys and used by the Swedish Hockey League (SHL), SmartCrop-H solves the unique challenges posed by ice hockey’s speed and the small size of the puck. This technology ensures that videos are optimized for various aspect ratios, making them perfect for sharing on today’s social media platforms.

Our solution relies on state-of-the-art AI models to detect and track the hockey puck, as well as players, to identify the most important regions of interest within each video frame. This guarantees that the action stays in focus, even on smaller screens or in non-standard aspect ratios like 1:1 and 9:16.

The Problem

Ice hockey is one of the most difficult sports to represent visually on digital platforms due to its fast movement, the size of the puck, and the complexity of tracking multiple players in real-time. Traditional broadcasting formats struggle to capture the excitement of the game when viewed on smaller or vertically-oriented screens, leading to a less immersive experience for fans on social media.

Manually cropping ice hockey videos to ensure the puck and players remain in focus is labor-intensive and often inaccurate, making it necessary to automate the process.

The Solution: AI-Powered Smart Cropping

SmartCrop-H tackles this problem by utilizing a cutting-edge AI pipeline that automatically tracks the puck and key players throughout a match. This AI system detects regions of interest in real-time, ensuring that the crop always centers on the most critical parts of the action.

The key components of the SmartCrop-H pipeline include: - Object detection: Focuses on identifying the puck and players during high-speed action. - Scene detection: Understands the overall context of the game to help frame the important moments. - Outlier detection: Filters out irrelevant background or sudden movements, ensuring smooth transitions. - Smoothing and interpolation: Ensures continuity and prevents jarring cuts or shifts during gameplay.

This approach allows dynamic cropping that adapts to different aspect ratios like square and vertical formats, ensuring high-quality video presentation regardless of the platform.

Key Features:

  • Puck and player tracking: Accurate detection of the small, fast-moving puck and key players.
  • Configurable aspect ratios: Optimized cropping for diverse formats, including 16:9, 1:1, and 9:16.
  • Real-time adjustments: Continuously updates the crop during gameplay to ensure seamless viewing.

Deployment and Impact

SmartCrop-H is already in production with Forzasys and is being used by teams in the Swedish Hockey League (SHL) to enhance their video content. This technology is revolutionizing how ice hockey is presented on social media and in mobile applications, ensuring that fans get a tailored, immersive experience.

Comprehensive evaluations of SmartCrop-H, both objective and subjective, have demonstrated its efficiency. A user study even showed that the output quality of SmartCrop-H is on par with, or exceeds, that of professional video editing tools, proving its value in automating the video cropping process without sacrificing quality.

Podcast

For a deeper dive into the SmartCrop-H project, listen to the podcast episode below, where we discuss the technical details and real-world applications of this innovative AI system:

Conclusion

SmartCrop-H is transforming how ice hockey is broadcasted and consumed across digital platforms. By leveraging advanced AI to detect and track the puck and players, SmartCrop-H delivers a high-quality, seamless viewing experience optimized for different aspect ratios. This technology is already having a significant impact on the SHL, bringing fans closer to the fast-paced action of ice hockey than ever before.

References

Sarkhoosh, M.H., Dorcheh, S.M.M., Midoglu, C., Sabet, S.S., Kupka, T., Johansen, D., Riegler, M.A. and Halvorsen, P., 2024. AI-Based Cropping of Ice Hockey Videos for Different Social Media Representations. IEEE Access.