Automated player identification and indexing using two-stage deep learning network.

Area of Science:

• Computer Vision
• Machine Learning
• Sports Analytics

Background:

• Accurate player identification and participation indexing are crucial for American football video analysis.
• Challenges in processing football video include crowded scenes, object distortion, and imbalanced data, particularly for jersey number recognition.

Purpose of the Study:

• To develop a deep learning-based system for automatic player tracking and participation indexing in American football games.
• To address the challenges of player identification in complex football video environments.

Main Methods:

• A two-stage deep learning network was designed for player detection and jersey number recognition.
• An object detection network (Detection Transformer) was used for player identification in crowded settings.
• A convolutional neural network was employed for jersey number recognition, synchronized with a game clock subsystem.

Main Results:

• The system accurately tracks players and identifies jersey numbers, enabling reliable play indexing.
• Qualitative and quantitative analyses demonstrated the effectiveness and reliability of the proposed player tracking system.
• The system successfully outputs a comprehensive log for play indexing in a database.

Conclusions:

• The developed deep learning system offers an effective solution for automatic player tracking and participation indexing in American football.
• The system shows significant potential for practical implementation in analyzing broadcast football videos.
• This technology can enhance the indexing and analysis of player performance in American football.