System and method for player reidentification in broadcast video
A system and method of re-identifying players in a broadcast video feed are provided herein. A computing system retrieves a broadcast video feed for a sporting event. The broadcast video feed includes a plurality of video frames. The computing system generates a plurality of tracks based on the plurality of video frames. Each track includes a plurality of image patches associated with at least one player. Each image patch of the plurality of image patches is a subset of the corresponding frame of the plurality of video frames. For each track, the computing system generates a gallery of image patches. A jersey number of each player is visible in each image patch of the gallery. The computing system matches, via a convolutional autoencoder, tracks across galleries. The computing system measures, via a neural network, a similarity score for each matched track and associates two tracks based on the measured similarity.
1. A method, comprising:
retrieving, by a computing system, a broadcast video feed for a sporting event, the broadcast video feed comprising a plurality of video frames;
generating, by the computing system, a plurality of tracks based on the plurality of video frames, wherein each track comprises a plurality of image patches associated with at least one player, each image patch of the plurality of image patches being a subset of a corresponding frame of the plurality of video frames;
connecting, by the computing system, gaps between sets of tracks of the plurality of tracks, the connecting comprising generating a motion field that estimates player motion across a set of video frames of the plurality of video frames;
constructing, by the computing system, a predicted motion for each player in the broadcast video feed based on the motion field for a respective player; and
generating, by the computing system, a graphical representation of the predicted motion for each player across a respective set of video frames.
2. The method of claim 1 , wherein retrieving, by the computing system, the broadcast video feed for the sporting event comprises:
generating a unified view of the broadcast video feed by removing portions of the broadcast video feed that are untrackable.
3. The method of claim 1 , wherein generating, by the computing system, the plurality of tracks based on the plurality of video frames comprises:
matching pairs of image patches from the plurality of image patches based on appearance and distance.
4. The method of claim 3 , wherein matching the pairs of image patches comprises:
measuring a color histogram similarity between two image patches to determine whether to match the two image patches as pairs.
5. The method of claim 1 , wherein the motion field is a vector field that measures a direction a player at a point on a playing surface will be after a predefined duration of time.
6. The method of claim 1 , wherein constructing, by the computing system, the predicted motion for each player in the broadcast video feed comprises:
predicting, via a neural network, player trajectories based on ground truth player trajectories and the motion field.
7. The method of claim 1 , wherein retrieving, by the computing system, the broadcast video feed for the sporting event comprises:
retrieving one or more sets of metadata associated with the plurality of video frames, the one or more sets of metadata comprising at least one of body pose information for each player and camera calibration data.
8. A non-transitory computer readable medium comprising one or more sequences of instructions, which, when executed by one or more processors, causes a computing system to perform operations, comprising:
retrieving, by the computing system, a broadcast video feed for a sporting event, the broadcast video feed comprising a plurality of video frames;
generating, by the computing system, a plurality of tracks based on the plurality of video frames, wherein each track comprises a plurality of image patches associated with at least one player, each image patch of the plurality of image patches being a subset of a corresponding frame of the plurality of video frames;
connecting, by the computing system, gaps between sets of tracks of the plurality of tracks, the connecting comprising generating a motion field that estimates player motion across a set of video frames of the plurality of video frames;
constructing, by the computing system, a predicted motion for each player in the broadcast video feed based on the motion field for a respective player; and
generating, by the computing system, a graphical representation of the predicted motion for each player across a respective set of video frames.
9. The non-transitory computer readable medium of claim 8 , wherein retrieving, by the computing system, the broadcast video feed for the sporting event comprises:
generating a unified view of the broadcast video feed by removing portions of the broadcast video feed that are untrackable.
10. The non-transitory computer readable medium of claim 8 , wherein generating, by the computing system, the plurality of tracks based on the plurality of video frames comprises:
matching pairs of image patches from the plurality of image patches based on appearance and distance.
11. The non-transitory computer readable medium of claim 10 , wherein matching the pairs of image patches comprises:
measuring a color histogram similarity between two image patches to determine whether to match the two image patches as pairs.
12. The non-transitory computer readable medium of claim 8 , wherein the motion field is a vector field that measures a direction a player at a point on a playing surface will be after a predefined duration of time.
13. The non-transitory computer readable medium of claim 8 , wherein constructing, by the computing system, the predicted motion for each player in the broadcast video feed comprises:
predicting, via a neural network, player trajectories based on ground truth player trajectories and the motion field.
14. The non-transitory computer readable medium of claim 8 , wherein retrieving, by the computing system, the broadcast video feed for the sporting event comprises:
retrieving one or more sets of metadata associated with the plurality of video frames, the one or more sets of metadata comprising at least one of body pose information for each player and camera calibration data.
15. A system comprising:
a processor; and
a memory having programming instructions stored thereon, which, when executed by the processor, causes the system to perform operations comprising:
retrieving a broadcast video feed for a sporting event, the broadcast video feed comprising a plurality of video frames;
generating a plurality of tracks based on the plurality of video frames, wherein each track comprises a plurality of image patches associated with at least one player, each image patch of the plurality of image patches being a subset of a corresponding frame of the plurality of video frames;
connecting gaps between sets of tracks of the plurality of tracks, the connecting comprising generating a motion field that estimates player motion across a set of video frames of the plurality of video frames;
constructing a predicted motion for each player in the broadcast video feed based on the motion field for a respective player; and
generating a graphical representation of the predicted motion for each player across a respective set of video frames.
16. The system of claim 15 , wherein retrieving the broadcast video feed for the sporting event comprises:
generating a unified view of the broadcast video feed by removing portions of the broadcast video feed that are untrackable.
17. The system of claim 15 , wherein generating the plurality of tracks based on the plurality of video frames comprises:
matching pairs of image patches from the plurality of image patches based on appearance and distance.
18. The system of claim 17 , wherein matching the pairs of image patches comprises:
measuring a color histogram similarity between two image patches to determine whether to match the two image patches as pairs.
19. The system of claim 15 , wherein constructing the predicted motion for each player in the broadcast video feed comprises:
predicting, via a neural network, player trajectories based on ground truth player trajectories and the motion field.
20. The system of claim 15 , wherein retrieving the broadcast video feed for the sporting event comprises:
retrieving one or more sets of metadata associated with the plurality of video frames, the one or more sets of metadata comprising at least one of body pose information for each player and camera calibration data.