Freeview video scaling in a real-world application
In this thesis, we want to make a system where the user may view a region of interest in a large panorama video. This means that the system should "zoom" in on the region and render the images in such a way that the extracted video has the correct perspective. This poses several challenges like smooth rendering, real-time processing and efficient delivery of large videos to client devices.
The scenario is based on a setup at Alfheim stadium in Tromsø. We have 5 2K-cameras generating a large panorama video and the idea is that a large number of users should be able to extract his own view from the point of the camera array making his own virtual camera. This eliminates the monotonous nature of the current broadcasting technology where the camera man and the crew decide on the region of interest. Instead, the freedom lies with the user to select a region of interest depending on various demands like focus on one player, on one specific team or on the defense of one team etc. The research tasks include evaluating different tradeoffs with respect to processing, data transfers and visual quality for the resulting video, and scaling the system to 1000s of concurrent users.
The topic is associated with the iAD centre for research based innovation where we have own labs at both IFI and Simula. In this particular project, we cooperate with Tromsø Idrettslag (TIL), ZXY sport tracking, Norsk Toppfotball/Cisco and the University of Tromsø. There is a running prototype (called Bagadus) used at Alfheim. The goal in this thesis is to build a real running prototype to be used by Norsk Toppfotball.
As there are several open issues, this topic can be shared by several students collaborating in the project.