Sber AI and CSKA Aim to Rewrite the Playbook for Sports Analytics
Sberbank's Sber AI team and Professional Football Club CSKA have developed a technology that can identify individual soccer players throughout an entire match using video from just a single camera. The approach could make advanced sports analytics significantly more affordable and accessible.
The technology is being viewed as a new step forward in evaluating both matches and player performance. By eliminating the need for multiple synchronized cameras, it delivers the data needed for in-depth analysis from a single video recording, reducing both complexity and cost.
International-Caliber Innovation
Modern soccer analytics relies on tracking data – precise information about the movements of players and the ball that is used to generate heat maps, calculate distance covered, analyze passing patterns, evaluate pressing, and understand team structure. Producing those insights requires a system capable of continuously tracking every player throughout a match using video footage.
Reliable data collection, however, is constantly challenged by real-world conditions. A player may leave the camera's field of view, become hidden behind other players, blend into an opponent during a contested play, or temporarily become visually indistinguishable. The system's task is to reacquire that player and connect fragmented observations into one continuous identity. The greatest challenge comes when the system mistakes one player for another. Once that happens, every subsequent statistic becomes unreliable, with data attributed to the wrong athlete and the resulting analysis painting a false picture of the match.
That challenge became the focus of the Sber AI team. Working together with Professional Football Club CSKA, the researchers became the first to define it as an independent scientific problem, naming it Long-Term Player Identification (LTPI). They analyzed a complete 101-minute match, assembled a dedicated dataset, and introduced a new evaluation method called the Cost-Sensitive Identification Score (CSIS), a metric that measures identification quality while accounting for the cost of different types of errors. The system identifies players by combining three independent signals: jersey number, team affiliation through kit color, and visual appearance, including height, body build, and movement patterns. The research describing the new technology was presented at the CVPR 2026 international conference and was selected as one of the Best Paper Award finalists in Denver, Colorado.
Making Elite Analytics More Accessible
Testing showed that the system confidently identified the correct player in 78% of cases. In the remaining 22%, it deliberately labeled the player as "undetermined" rather than risking a misleading identification. Its most significant advantage, however, is that it extracts high-value tracking data from a standard broadcast feed without requiring expensive specialized hardware.
"Together with Professional Football Club CSKA, we have created a foundational technology that opens the door to the future of 'invisible scouting' – an AI-powered system capable of analyzing matches using ordinary video and single-camera tracking. This approach enables a gradual shift from manual observation to scalable analytics by tracking player movement across the entire field, evaluating performance over time, identifying consistent progress, and highlighting areas where greater intensity and higher-quality work are needed. Our solution delivers value on three levels," said Semyon Budyonny, Managing Director and Head of Advanced Technology Development at Sberbank.
The technology developed by the Russian team could benefit sports technology companies, analytics platforms, professional clubs, scouts, and providers of video tracking solutions. Its applications extend well beyond soccer to other sports, while also making advanced performance analysis available to teams in smaller cities and supporting the development of young athletes in environments where sophisticated analytics has traditionally been out of reach because of cost.
Where the Technology Goes Next
The Sber AI and CSKA project continues the broader shift in sports analytics away from expensive integrated hardware-and-software systems and toward software that can work with conventional video recordings. The first users are expected to include football academies, regional clubs, and companies building analytics services for organizations that cannot justify the cost of traditional tracking systems.
Before large-scale deployment, the technology will require larger training datasets and broader validation across many more matches, camera angles, weather conditions, and uniform variations. Looking ahead, the method could become part of a comprehensive platform capable of automatically analyzing not only player movement but also technical execution, tactical formations, and long-term player development. Russian researchers have created a technology that could ultimately serve as the foundation for a new generation of advanced sports analytics.
