Method and system of automatically estimating a ball carrier in team sports

1 . A computer-implemented method comprising:

obtaining three-dimensional (3D) location data of individual players and a ball based on images of an athletic event on a field;

generating feature vectors corresponding respectively to the individual players on the field, the feature vectors including amounts of other individual players in neighbor grid positions relative to respective ones of the individual players, a first one of the feature vectors for a first one of the individual players including an amount of the other individual players in a five-by-five space grid with the first one of the individual players at a center of the five-by-five space grid; and

automatically estimating a ball carrier position on the field based on inputting the feature vectors into a convolutional neural network, the convolutional neural network trained to output a ball carrier probability map including probabilities of a ball carrier being at different locations on the field.

2 . The method of claim 1 , wherein the feature vectors are arranged into channels having different one or more features than other channels.

3 . The method of claim 2 , wherein the channels have a size corresponding to a player position grid, the player position grid corresponding to a size of the field, and the individual players are assigned to grid spaces of the player position grid.

4 . The method of claim 1 , wherein the feature vectors include 3D positions of the individual players.

5 . The method of claim 1 , wherein the feature vectors include at least one of velocities or directions of movement of the individual players.

6 . The method of claim 1 , wherein the feature vectors include orientations of the individual players.

7 . A computer-implemented system comprising:

memory storing at least images of an athletic event on a field, the images from multiple cameras;

instructions; and

processor circuitry communicatively coupled to the memory, the processor circuitry to be programmed based on the instructions to:

generate three-dimensional (3D) location data of individual players and a ball from images;

generate feature vectors corresponding respectively to the individual players on the field, the feature vectors including amounts of other individual players in neighbor grid positions relative to respective ones of the individual players, a first one of the feature vectors for a first one of the individual players including an amount of the other individual players in a five-by-five space grid with the first one of the individual players at a center of the five-by-five space grid; and

input the feature vectors into a convolutional neural network to estimate a position of a ball carrier on the field, the convolutional neural network trained to output a ball carrier probability map including probabilities of the ball carrier being at different locations on the field.

8 . The system of claim 7 , wherein the feature vectors include data related to player attributes, movement and location of the players, a ball position relative to the players, and multiple frame-related data.

9 . The system of claim 7 , wherein the first one of the feature vectors includes a count of neighbor players on a same team as the first one of the individual players and a count of players not on the same team as the first one of the individual players.

10 . At least one non-transitory computer readable medium comprising instructions to cause a computing device to at least:

obtain three-dimensional (3D) location data of individual players and a ball from images of an athletic event on a field;

generate feature vectors corresponding respectively to the individual players on the field, respective ones of the feature vectors including an amount of the other individual players in a five-by-five space grid with the corresponding one of the individual players at a center of the five-by-five space grid; and

input the feature vectors into a convolutional neural network to estimate a position of a ball carrier on the field, the convolutional neural network trained to output a ball carrier probability map including probabilities of the ball carrier being at different locations on the field.

11 . The non-transitory computer readable medium of claim 10 , wherein the feature vectors comprise multi-frame related data indicating differences from one frame to another frame.

12 . The non-transitory computer readable medium of claim 10 , wherein the feature vectors are arranged into channels having different one or more features than other channels.

13 . The non-transitory computer readable medium of claim 10 , wherein the instructions are to cause the computing device to determine candidate players based on the ball carrier probability map.

14 . The non-transitory computer readable medium of claim 10 , wherein the feature vectors include at least one of velocities or directions of movement of the individual players.

15 . The non-transitory computer readable medium of claim 10 , wherein the feature vectors include orientations of the individual players.

16 . The non-transitory computer readable medium of claim 10 , wherein the feature vectors include at least one of a player jersey number, player motion and position, a ball position, or at least one difference in at least one of the ball or at least one player from one frame to another frame.