The National Basketball Association (NBA) used to provide high-fidelity tracking data for its games. Every 0.04 seconds, high-speed cameras would report cartesian coordinates for the ball and the 10 players on the court. From this data, we can generate videos/animations that show a top-down view of an entire basketball game. This final capstone project uses various deep learning techniques to attempt to answer two questions: 1) Can we identify scoring events (a made 2-point shot, missed 2-point shot, missed 3-point shot, made 3-point shot) from these videos? 2) Can we "predict" what the result of a sequence or play is by showing our deep learning model a slice of video leading up to the actual slicing event? Come watch if you have any interest in deep learning, sports, video action recognition, etc.
No comments yet