Building a Dataset and Developing a Video Event Classifier for Football

The challenges and inaccuracies from manually collecting and processing event data for football have highlighted an increasing need to automate event detection. While leveraging tracking data makes it possible to begin extracting events automatically, it becomes difficult to differentiate between events which share a similar context, such as types of duels, saves, fouls, stoppages, and restarts. Video classification, a well-established computer vision tool for identifying events in video clips, can be used in applications where tracking data alone fails to retell the game in its entirety. In this paper, we develop an end-to-end video classification pipeline to identify player duels in football using data from the 2022 Qatar Men’s World Cup. The methodology includes syncing manually annotated events with game video, generating 3 second video clips for all duel-like events in the tournament, and fine-tuning pretrained 3D convolutional neural networks to produce event predictions. We conduct several experiments to compare various camera angles, video resolutions, and binary versus multi-class models. We find that binary models outperform multi-class models significantly. To further improve the performance, future iterations can optimize the training parameters and increase the number of examples to narrow this gap.

Name thesis:degree_name

Master

Department dc:contributor.department

Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2023

Author dc:creator

• Best Jr., Reginald

Advisors dc:contributor.advisor

• Chase, Christina
• Vidal-Codina, Ferran

Statement dc:rights

• In Copyright - Educational Use Permitted
• Copyright retained by author(s)

Licence dc:rights.uri

• https://rightsstatements.org/page/InC-EDU/1.0/

Handle dc:identifier.uri

https://hdl.handle.net/1721.1/151510

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/151510