Broadcast Video Player Tracking for Cross-League Rating Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Professional sports teams lack accurate tracking data for non-professional league players, limiting their ability to make informed decisions in draft selection and performance prediction, due to the high cost and limited adoption of optical player and ball tracking systems in non-professional leagues.
Innovation Solution
Utilizing state-of-the-art computer vision techniques to capture player and ball tracking data from broadcast video, generating detailed tracking data for non-professional leagues, and applying machine learning models to predict player performance in professional leagues based on this data, including techniques for generating daily-updated performance ratings (DRIP) and Wins Above Replacement (WAR) metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical player and ball tracking systems are deployed in non-professional leagues, then measurement precision of player performance data is improved, but device complexity and cost increase significantly
Solution Approach 1:
The patent uses computer vision to create a digital copy of the tracking data that would be produced by physical optical systems. Instead of deploying actual tracking cameras in non-professional leagues, the system processes broadcast video to generate equivalent tracking information, thereby achieving measurement precision without the device complexity and cost of physical tracking infrastructure
Solution Approach 2:
The patent replaces the mechanical/optical tracking system with a computational approach using computer vision and machine learning. The system substitutes physical tracking infrastructure with algorithms that extract player and ball positions from broadcast video, eliminating the need for complex optical hardware while maintaining tracking data quality
2Loss of information
If broadcast video is processed to generate tracking data, then availability of performance data in non-professional leagues is improved, but processing time and computational resources increase
Solution Approach 1:
The system performs preliminary processing by pre-extracting player and ball tracking data from broadcast videos and storing them in structured formats. This advance preparation allows the data to be readily available for subsequent analysis without requiring time-consuming processing at the point of use, thereby reducing loss of time while improving data availability
Solution Approach 2:
The system creates self-contained tracking data files that can be independently processed and analyzed without requiring the original broadcast video or additional processing resources. The extracted tracking data serves itself as a complete data source, eliminating repeated processing needs and reducing computational overhead
3Adaptability or versatility
If machine learning models are trained on tracking data from multiple leagues, then adaptability of performance prediction is improved, but training data requirements and model complexity increase
Solution Approach 1:
The patent transforms tracking data from different leagues into a unified parameter space by normalizing player positions, movements, and game contexts. This parameter transformation allows models trained on one league's data to adapt to other leagues without requiring extensive additional training data, thereby improving league compatibility while managing training data requirements
Solution Approach 2:
The system develops machine learning models with universal applicability across multiple leagues by designing them to handle varied playing styles, court dimensions, and game rules. The models are constructed to process tracking data from any league in a standardized manner, enabling one model to serve multiple functions across different basketball leagues without requiring separate models for each league
Data Source
AI summary
Disclosed techniques relate to utilizing tracking data for predicting player ratings. In an example, a method for utilizing tracking data to predict a player rating includes receiving broadcast data for a plurality of games in a first league, the plurality of games including a first player, generating tracking data for each of the plurality of games, the tracking data comprising coordinates of player positions and ball positions for each frame of the broadcast data, receiving play-by-play data for each of the plurality of games, the play-by-play data describing events that occur within the plurality of games, merging the tracking data and play-by-play data to generate a set of input features, and predicting, based on the set of input features, a player rating for the first player, the player rating being indicative of a predicted level of performance in a second league.