A Topic Model Approach to Representing and Classifying Football Plays

We address the problem of modeling and classifying American Football offense\udteams’ plays in video, a challenging example of group activity analysis. Automatic play\udclassification will allow coaches to infer patterns and tendencies of opponents more ef-\udficiently, resulting in better strategy planning in a game. We define a football play as a\udunique combination of player trajectories. To this end, we develop a framework that uses\udplayer trajectories as inputs to MedLDA, a supervised topic model. The joint maximiza-\udtion of both likelihood and inter-class margins of MedLDA in learning the topics allows\udus to learn semantically meaningful play type templates, as well as, classify different\udplay types with 70% average accuracy. Furthermore, this method is extended to analyze\udindividual player roles in classifying each play type. We validate our method on a large\uddataset comprising 271 play clips from real-world football games, which will be made\udpublicly available for future comparisons