System and method for predictive sports analytics using body-pose information

Patent metadata

Field	Value
Publication number	US-10824918-B2
Application number	US-201815885668-A
Country	US
Kind code	B2
Filing date	Jan 31, 2018
Priority date	Jan 31, 2017
Publication date	Nov 3, 2020
Grant date	Nov 3, 2020

Abstract

A system is described for analyzing plays of a sporting event based on body-pose information in conjunction with real-world positional tracking data. The system permits segmentation of sporting event plays into discrete time periods and the labeling of appropriate body-pose characteristics for each period. The system allows for comparison and fine-grained analysis of plays with respect to either a series of body-pose vertices that exist in continuous space (i.e., skeleton) or map directly to attributes, using the positional tracking data and play information to account for contextual differences. To enable analysis for the former, the system performs a projection to 3D space, followed by a spatiotemporal alignment step. Through the system, the importance of particular body-pose motions or specific attributes to the success of particular sporting event plays is quantified.

First claim

The invention claimed is:

1. A computer-implemented sports analytics system comprising:
a database storing a plurality of plays for a given sporting event, each play comprising real-world positional information and game event information over a window of time for objects on a playing surface, the objects comprising a plurality of players and a ball;
a body-pose extraction system for extracting body-pose information for each player of the plurality of players from a video broadcast of the sporting event; and
a computing device, coupled to the database and the extracted body-pose information, for assigning values to a set of body-pose attributes for plays in the database, the computing device operable to:
receive a selection of a set of plays from the sporting event; and
statistically compare the selected set of plays to a second set of plays with respect to the assigned body-pose attribute values for each player of the plurality of players involved in the play.

2. The system of claim 1 , the computing device further comprising a display unit, and further operable to display an output of the statistical comparison on the display unit in a form of a histogram of body-pose attributes.

3. The system of claim 1 wherein the given sporting event is a basketball game and the plays are three-point shot attempts.

4. The system of claim 1 wherein the set of body-pose attributes is inferred from the set of plays using a clustering algorithm.

5. A computer-based method for analyzing body-pose information for a set of plays in a sporting event, the set of plays represented in a database including positional information for the set of plays, the method comprising:
obtaining body pose information for the set of plays, the body pose information corresponding to each player of a plurality of players involved in the set of plays;
segmenting each play into two or more discrete segments and selecting a set of body pose attributes for each segment;
evaluating the body pose attributes for each segment of each play in the set of plays and assigning values for each attribute;
training a play classifier using the positional information; and
analyzing the set of plays using the body-pose information and the play classifier.

6. The method of claim 5 , wherein training the play classifier comprises:
retrieving positional information and play-outcome information for the set of plays using the positional information from the database;
generating spatial and temporal data for the set of plays from the positional information;
selecting spatial and temporal features for evaluation;
correlating values of the selected spatial and temporal features for the set of plays with the play-outcome information for the set of plays to generate a likelihood of success for each play; and
classifying the plays according to their likelihood of success.

7. The method of claim 5 , wherein analyzing the set of plays using the body-pose information and the play classifier comprises:
applying a Pearson's chi-squared test to compare body-pose attribute values across two sets of classified plays.

8. The method of claim 7 , further comprising:
generating a histogram for display, the histogram illustrating differences in body-pose attribute values between the two sets of classified plays.

9. The method of claim 5 , wherein the body-pose information is obtained from a monocular view from a broadcast video of the sporting event.

10. The method of claim 5 , wherein the sporting event is basketball.

11. The method of claim 10 , wherein the set of plays comprises three-point shot attempts.

12. The method of claim 5 , wherein the selecting of body pose attributes comprises:
estimating 3D positions from 2D positional information; and
aligning the body-pose information spatially and temporally using the estimated 3D positions.

Assignees

• Stats Llc

Inventors

• Felsen Panna
• Lucey Patrick

Classifications

• G06F18/2415

  based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06F18/23

  Clustering techniques · CPC title

• G06V40/23

  Recognition of whole body movements, e.g. for sport training · CPC title