Generating Visual Analytics And Player Statistics For Soccer

Patent Status

Full Description

Background

Identification of next generation sports stars is an important responsibility of a coach. Talent identification has been traditionally based on viewing athletes in a trial game or training session environment. A coach's subjective preconceived notion of the ideal player may result in misjudgments and inconsistencies. In team-based sports, such as soccer, talent identification is a complex process due to different qualities associated with performance including personal and tactical attributes.

Researchers led by Prof. Bir Bhanu at UCR have designed a patent pending system to automate talent identification by generating visual analytics and player statistics for soccer from a video using traditional machine learning algorithms and deep learning techniques for computer vision. Specifically, they have developed:

• An approach to generate player analytics and statistics from videos of soccer matches.
• Convolutional Neural Networks for dynamic identification of players controlling the ball.
• Strategy to train Generative Adversarial Networks to augment and improve the performance of the system.
• Generalizable approach for use during different scenarios of the game.

Example scenarios of players with and without the ball

Sample image of the grid-based localization technique used in the invention

Advantages

The system and approach that the inventors have developed, now provides:

• An approach to generate player analytics and statistics from videos of soccer matches.
• Convolutional Neural Networks for dynamic identification of players controlling the ball.
• Strategy to train Generative Adversarial Networks to augment and improve the performance of the system.
• Generalizable approach for use during different scenarios of the game.

Suggested uses

• Video analytics in sports
• Talent identification in sports
• Player development in sports
• Game strategy development in Soccer

State Of Development

Proof of concept prototype developed and tested. The testing displays an impressive 92.57% ± 2.92% accuracy in identifying teams. For player analytics their accuracies were, in each case:

• 84.73% - for easy scenarios - 4 – 5 players spread wide apart, e.g., in the defense zone.
• 79.82% - for moderate scenarios - 6 – 10 players in the midfield.
• 67.28% - for hard scenarios - more than 10 players in a small area - e.g., during an attempt at a goal.

Inventions by Bir Bhanu

Inventions by Bir Bhanu 

Related Materials

• Soccer: Who Has The Ball? Generating Visual Analytics and Player Statistics