Automatic video event clipping in sports

Use machine learning and video processing techniques to cut video events in sports videos. E.g., find the best start and stop times for a goal in a soccer video.
Master

Sports events are currently manually tagged by human operators in a tedious process. To enable fast and efficient distribution of soccer event (can be video in general), we have developed a system where an operator just press a button for the type of event (like a goal) and publish. In this case, the event is available really fast - faster than the broadcast replays, but it is inaccurate cut (and therefore require a second process of cutting. The idea of this task is to use machine learning and video processing in order to find the scene cuts and give the best possible time interval for the video footage.

To see examples of several non-optimal clippings of events, see the current highlights page for Eliteserien in Norway: highlights.eliteserien.no

We will be using real videos from this site for the thesis.

Goal

A system (or machine learning model) that can assist the annotation (production) of video events in a video streaming service running for the Norwegian and Swedish elite leagues in soccer.

Learning outcome

Video processing, machine learning, real-world implementation and experiments

Qualifications

  • Python programming
  • Knowledge about deep learning and video processing is an advantage

Supervisors

  • Pål Halvorsen
  • Michael Riegler
  • Tomas Kupka
  • Ragnar Langseth

Collaboration partners

ForzaSys AS

References

highlights.eliteserien.no

Contact person