Video is a rich source for motion analytics; however, many analyses of video are infeasible due to high computational complexity. This project aims to improve human-in-the-loop analysis of video for the application of soccer. We expect that augmenting user annotations to existing tracking algorithms will improve both accuracy and efficiency. We will look to procure and leverage sparse, high-quality user annotations to resolve instances of high representation uncertainty in player trackers. Probabilistic graphical models, a key component of our research approach, provide the uncertainty quantification and interpretability needed to accomplish this goal.

“As both a computer scientist and a soccer player, I am fascinated by this project because it tries to apply novel computational analyses to an age-old challenge for soccer coaches and players alike: analyzing the strategy of a soccer match. This research may also act as a springboard for future graduate research and my career post-MIT.”