See slides here
Football coaches spend countless hours “breaking down game film” to best prepare their team for success; discovering a pattern may be the difference between victory and defeat. One fundamental element of game film analysis is recording an opponent’s playbook – this process is currently done manually if at all. We have automated and incorporated this and other film analysis tasks within our prototype, “AutoScout.”
In this presentation, we discover playbooks and learn to classify play-types automatically using supervised topic models, a powerful data mining framework. Play-type patterns are determined by distinct motion directions and routes taken by different players.
Preliminary experiments were performed on a dataset of 271 play clips from real-world football games. Classification results indicate that our method can predict play-types with ~80% accuracy.