Events

Measuring team and player performance in soccer is extremely challenging due to the continuous, low-scoring and strategic nature of the game. Couple these issues with the fact that the “world game” has so many different leagues and competitions around the globe which vary in terms of quality, style, and data coverage – this problem is compounded. In this talk, we show how using machine learning and computer vision can get around these issues to objectively measure team and player performance. Specifically, using an enormous amount of tracking, event and video data we show: i) how we can forecast how a player will play on another team, ii) how to accurately measure formations in soccer, and iii) how to capture tracking data from broadcast video.