Optimal play route prediction during sports broadcasts

Patent metadata

Field	Value
Publication number	US-12547801-B1
Application number	US-202418973924-A
Country	US
Kind code	B1
Filing date	Dec 9, 2024
Priority date	Dec 9, 2024
Publication date	Feb 10, 2026
Grant date	Feb 10, 2026

Abstract

Systems and techniques are described for providing optimal play route prediction during sports broadcasts. In various examples, a first frame of tracking data indicating a first respective location of each player of a first plurality of players on a field plane may be received at a first time. A first predicted direction of motion may be determined for the first frame based at least in part on a first objective. A second frame of tracking data indicating a second respective location of each player of the first plurality of players on the field plane may be received at a second time. A second predicted direction of motion for the second frame may be determined based at least in part on the first objective. A first graphical overlay may be displayed on a live video feed based on one or more of the first predicted direction and the second predicted direction.

First claim

What is claimed is:

1 . A computer-implemented method comprising:
receiving a first frame of tracking data indicating a first respective location of each player of a first plurality of players on a field plane at a first time, wherein the first frame of tracking data comprises data generated by a first plurality of sensors; executing a first simulation using a physics-based simulator and the first frame of tracking data, wherein the first simulation uses a set of sport-specific constraints to solve an optimization problem; generating, by the first simulation using the first frame of tracking data, first path data indicating a path for a first player of the first plurality of players, wherein the first path is determined by the first simulation by maximizing an objective associated with the first simulation; receiving a second frame of tracking data indicating a second respective location of each player of the first plurality of players on the field plane at a second time after the first time, wherein the second frame of tracking data is generated by the first plurality of sensors; executing a second simulation using the physics-based simulator and the second frame of tracking data, wherein the second simulation uses the set of sport-specific constraints to solve the optimization problem; generating, by the second simulation using the second frame of tracking data, second path data indicating a modified path for the first player of the first plurality of players relative to the first path data, wherein the second path is determined by the second simulation by maximizing the objective associated with the second simulation; and causing a first graphical overlay representing the second path data to be displayed on a live video feed depicting at least the first player.

2 . The computer-implemented method of claim 1 , further comprising:
determining, using a third frame of tracking data and a transformer-based classifier, a first predicted momentary direction of motion in the field plane, wherein the first predicted momentary direction of motion is determined by the transformer-based classifier to maximize a number of yards gained; and causing a second graphical overlay representing the first predicted momentary direction of motion to be displayed on the live video feed.

3 . The computer-implemented method of claim 2 , further comprising:
determining, using a fourth frame of tracking data and the transformer-based classifier, a second predicted momentary direction of motion in the field plane, wherein the second predicted momentary direction of motion is different from the first momentary direction of motion; and determining, over a first set of frames of tracking data, that a majority of frames of the first set of frames of tracking data are associated with the first predicted momentary direction of motion, wherein the first set of frames comprises the third frame and the fourth frame, wherein the second graphical overlay is caused to be displayed on the live video feed in response to the majority of frames being associated with the first predicted momentary direction.

4 . A method comprising:
receiving a first frame of tracking data indicating a first respective location of each player of a first plurality of players on a field plane at a first time, wherein the first frame of tracking data comprises data generated by a first plurality of sensors; determining, for at least the first frame of tracking data, a first predicted direction of motion based at least in part on a first objective; receiving a second frame of tracking data indicating a second respective location of each player of the first plurality of players on the field plane at a second time, wherein the second frame of tracking data is generated by the first plurality of sensors; determining, for at least the second frame of tracking data, a second predicted direction of motion based at least in part on the first objective; and causing a first graphical overlay to be displayed on a live video feed, wherein the first graphical overlay is based on one or more of the first predicted direction and the second predicted direction.

5 . The method of claim 4 , further comprising:
generating, using the first frame of tracking data and an encoder, first embedding data; and determining, by a supervised classifier model, the first predicted direction of movement.

6 . The method of claim 5 , wherein the supervised classifier model is trained to select a direction among a predefined set of angular directions that maximizes an objective associated with a first sport.

7 . The method of claim 5 , wherein the first embedding data comprises respective indicator data for each player of the first plurality of players, the respective indicator data indicating that the respective player is on an offensive team or a defensive team.

8 . The method of claim 4 , further comprising:
determining, over a first set of frames of tracking data comprising the first frame and the second frame, that a majority of frames of the first set of frames of tracking data are associated with the first predicted direction of motion, where the first graphical overlay indicates the first predicted direction of motion.

9 . The method of claim 4 , further comprising:
executing a first simulation using a physics-based simulator and the first frame of tracking data, wherein the first simulation uses a set of sport-specific constraints to solve an optimization problem, wherein the determining the first predicted direction of motion comprises determining a first path of motion in the field plane; and executing a second simulation using the physics-based simulator and the second frame of tracking data, wherein the second simulation uses the set of sport-specific constraints to solve the optimization problem, wherein the determining the second predicted direction of motion comprises determining a second path of motion in the field plane.

10 . The method of claim 9 , wherein the first frame of tracking data and the second frame of tracking data are sampled from first tracking data representing a single play.

11 . The method of claim 9 , wherein the first set of sport-specific constraints comprise:
first constraint data indicating that a player with a ball avoids defensive players; second constraint data indicating that the defensive players attempt to intersect with the player with the ball in the field plane; third constraint data indicating that the defensive players attempt to avoid offensive players other than the player with the ball; and fourth constraint data indicating that offensive players other than the player with the ball attempt to intersect with defensive players in the field plane.

12 . The method of claim 4 , wherein the first objective is to maximize a number of yards gained.

13 . A system comprising:
at least one processor; and non-transitory computer-readable memory storing instructions that, when executed by the at least one processor, are effective to perform operations comprising:
receiving a first frame of tracking data indicating a first respective location of each player of a first plurality of players on a field plane at a first time, wherein the first frame of tracking data comprises data generated by a first plurality of sensors;
determining, for at least the first frame of tracking data, a first predicted direction of motion based at least in part on a first objective;
receiving a second frame of tracking data indicating a second respective location of each player of the first plurality of players on the field plane at a second time, wherein the second frame of track

Assignees

• Amazon Tech Inc

Inventors

• Karklinsky Matan
• Pragier Gabriel
• Segev Bar
• Schwartzstein Sam
• Pawlowski Dan
• Rotman Michael
• Ben-Cohen Avi Avraham
• Ideses Ianir
• Zvik Yochai

Classifications

• H04N21/2187

  Live feed · CPC title

• H04N21/8146

  involving graphical data, e.g. 3D object, 2D graphics · CPC title

• G06F30/27 Primary

  using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model · CPC title

• A63B24/0021

  Tracking a path or terminating locations · CPC title

• A63B2024/0056

  for statistical or strategic analysis · CPC title