A domestic research team has developed artificial intelligence (AI) technology that automatically extracts and records key information about games, such as player positions, using only smartphone footage. This technology won an international soccer AI model competition. If commercialized, it is expected to enable game analysis at a professional sports level for local tournaments or school sports teams without expensive equipment. The 'AI Recorder' developed by the research team will be piloted in domestic amateur volleyball from this year.

The team led by Kim Jin-wook, Principal Researcher at the Korea Institute of Science and Technology (KIST) Center for Intelligence and Interaction Research, won first place in the 'SoccerNet Challenge Game Situation Reconstruction (GSR)' category at the 'Computer Vision and AI Conference (CVPR) 2025' held in the United States in June. In mid-month, Kim stated, "We originally started research with volleyball game analysis," adding, "We applied it to soccer, the sport with the highest demand for game analysis worldwide, and won first place in our first competition entry."

● Accurate Game Recording and Analysis Without Specialized Equipment

The SoccerNet Challenge is a competition at the forefront of soccer video-based AI technology. In its fifth year, the competition saw participation from a total of 75 teams from universities, research institutions, and companies worldwide. The GSR category aims to automatically reconstruct the state of the field over time during a game. Teams compete by submitting AI-generated answers based on the video provided, which includes player information and movement trajectories.

The research team optimized the tilted view of the soccer field and player positions captured from the side and improved tracking methods following player movements to secure first place. They implemented an AI model capable of precisely analyzing major events, player movements, and game situations using only low-resolution smartphone footage.

Game analysis data is essential for strategy development, player evaluation, and feedback. Before digital technology, game recorders relied on manually recorded observations. With the advancement of cameras and sensors, automation became possible. In professional soccer, players often wear tracker equipment under their uniforms to track positions.

Until now, obtaining game analysis data required installing numerous expensive cameras in the stadium and equipping each player with wearable devices. Kim stated, "While this is possible in professional and elite sports, it is quite challenging to apply such equipment in amateur, youth, or school sports," adding, "Winning this competition demonstrated that the AI model can replace or surpass existing systems' analysis performance and be practically utilized in resource-limited settings."

Data recorded by AI can assist referees, provide in-depth analysis to prevent injuries through player movement analysis, help formulate new strategies, and serve as a scouting basis to identify promising talents.

Recently, amateurs also record their games individually and often upload them to video platforms, continuously generating analyzable data. Kim stated, "When combined with generative AI, it can be turned into content such as automatic commentary and highlight generation," adding, "Just like baseball, where continuous data like batting averages or earned run averages engage fans, the data itself can offer new experiences to fans."

● “Expanding to Various Sports”

The developed technology is expected to expand to various sports disciplines. It can elevate the overall infrastructure level even in minor sports with limited demand and personnel.

In fact, the research team has contracted with the Korea Volleyball Association to pilot AI-based game record automation in amateur volleyball tournaments from this year and plans to officially launch the service next year. In domestic tennis, a pilot operation is planned for 2026, with official service starting in 2027. Related sales are expected to grow from KRW 150 million next year to KRW 800 million by 2029.

Kim stated, "I believe the core function of AI is to provide high-quality services at a low cost so that many people can enjoy them."

Popular News