Automatic Football Video Segmentation

This was a group project for CS6601 (Artificial Intelligence) for which I was the team lead. Group Members include: Ashwin Shenoi, Takahiko Tsuchiya, Bi Ge and Yue Liu.

This system performs an automatic play-by-play segmentation of college football videos using the audio information present in the video. The dataset consisted of four of Georgia Tech’s home football games collected by the eStadium VIP team at Georgia Tech.

The system was created in two stages: HMM based classifiers for plays and non-plays were trained using 1s audio clips extracted from the training set. GMM based clustering was performed to generate training labels. In the second system, game specific domain knowledge was incorporated into the system to prune the results of the first stage thereby increasing the precision of the system.

My responsibilities included: Algorithm design, Data Visualization and Project Scheduling and Organizing.

Final Poster



Leave a Reply

Your email address will not be published. Required fields are marked *