Object tracking for basketball analysis

[molten orbit]

Hi,

I am building a project where I have 2 cameras filming half of a basketball court (the same side of the court). One camera is filming from the right side and the other is filming from the left side.

The objects (players in this case) are present in the two cameras at the same time and my goal is to track the players and record the shots that go into the basket and the shots that fail.

So when the game is finished, I need to have a file (e.g. CSV) that displays the stats of the game.

for example:

player1, 5 shots scored, 2 shots missed, image1_player_1, image2_player_1, ...
player2, 7 shots scored, 5 shots missed, image1_player_2, image2_player_2, ...
player3, 1 shot scored, 2 shots missed, image1_player_3, image2_player_3, ...
...

The problem with tracking is re-ID and ID switching so I want to know what is the best way to not confuse players with each other?

For example, when player1 shoots the ball how do I actually know that this is player1? what technique do I have to use?

Can I use feature matching (SIFT, ORB, ...) in this case? Or maybe I would need to use some sort of clustering algorithm to cluster similar images (same player) together?

please guide me in the right direction so that I know how to do it! Thanks

[thorny trellis]

I don't work with CV so don't take my word for it but the simplest solution to your problem would be that you do detection on the 2 video feeds seperately and then decide the final result based on confidence scores.

Also this might be of your interest
https://www.marktechpost.com/2023/04/27/sustech-vip-lab-proposes-track-anything-model-tam-that-achieves-high-performance-interactive-tracking-and-segmentation-in-videos/