AI-based Video Analysis and Processing for Sports (multiple topics)
Interested in AI and video technology? In sports tech used by elite leagues and live broadcasters? If either, read on ...
We have for long researched sports technology, both for distributing the video of live broadcasts and managing the archives in Norway and Sweden. We develop systems, from camera capture to dissemination, that are actively used and tested on real users, where the users can see highlights and summaries, get recommendations, collect favorite events, make personalized playlists, etc. However, there are a lot of open questions and issues to address like how to make (usually automated, AI-based) that can be focused on (see examples under the goal section) where each of these points can be separate master topics or can be combined. Or do you have a related idea yourself that can help improve current video systems? We collaborate with the leagues and elite clubs in Norway and Sweden, the football associations, Telenor, broadcasters, Forzasys (a streaming technology company), Telenor and several international partners. In addition, we collaborate with the Swedish Hockey League, and are entering sports like basketball and tennis. Want to be part of our research group and improve the current systems? Contact us, and let's see if we can find a thesis topic that matches your interests!
Goals
There are many possible topics and goals for this area, for example:
- Event detection systems — can we locate where a particular event is in the video, understand the semantics, e.g., detect a goal, a booking, a shot, a drible, a tackle, etc.
- Clipping systems — find the best start and stop for a video clip of an event, e.g., find good clipping points for example not in the midle of a replay.
- Object tracking — for many tasks, it is important to track object like players, ball/puck, referee, sponsor logos, etc. For example, you want to clip a video tracking the ball or a particular player, or you want to show a sponsor how often their logo has been exposed to the viewers.
- Video shot classification — classify the different shot and camera angles, e.g., is this the first goal sequence, a replay, players cheering, a zoomed-out, the fans, etc. Such information is important when summarizing and clipping a video sequence consisting of multiple camera shots.
- Video and/or text summarization systems — how can we create a compilation of the main events of a video and create a highlight package like shown on the sport news, e.g., make automatic game highlight for the news or automatically write a blog post.
- Cropping systems — videos are usually recorded in a wide-screen format by the broadcasters targeting traditional TV devices, but people today often consume video on their phones on portrait format (standing, social media apps), and a question is how to capture the essence of a video to fit the new screen format, e.g., resize images and videos to match target devices and applications, for example going from a 16:9 TV format to a 1:1 Instagram format.
- Video search engines — video sites like YouTube contain a lot of information, but how can you find a particular single event. For example, football sites usually have 10.000s of goals. Can we make a search engine based on both existing metadata and spoken audio in the videos to make an efficient search?
Learning outcome
- Video analysis
- Machine learning / AI
- User studies
- Building real systems
Qualifications
- Hard working, motivated
- Some programming
- Interested in learning (the rest can be learned during the thesis work)
Supervisors
- Michael Riegler
- Pål Halvorsen
- Steven Hicks
Collaboration partners
- We have many collaboration partners, including
- Forzasys AS - https://forzasys.com
- Norsk Toppfotball
- Svensk Elitfotboll
- Svenske Hockeyligaen