AI Learns to Track Individual Soccer Players Using a Single Camera
Researchers from Sber AI and PFC CSKA have developed a system that delivers detailed player analytics without requiring expensive multi-camera installations.
During a live match, computer vision systems frequently lose track of players as they move out of the camera's field of view or become obscured while contesting the ball. Researchers from Sber AI and PFC CSKA have become the first to define the challenge of continuously tracking the same player throughout an entire match as a distinct scientific problem, which they call long-term player identification.
To evaluate the system's performance, the researchers developed a new metric called CSIS, which identifies players using jersey numbers, uniform colors, and visual appearance. The system correctly identified players in 78% of cases, while in the remaining 22% it explicitly classified them as "unknown" rather than making an incorrect prediction. The research was presented at the CVPR 2026 conference and was selected as a Best Paper Award recipient.
