Tables can be transformed into football pitches, hosting real 3D matches thanks to an augmented reality system.
Researchers from University of Washington, Facebook, and Google claim to have developed a learning based system that transforms a video of a football match into a moving 3D reconstruction that can be watched through a 3D viewer or augmented reality device.
The system is based on an approach that estimates the depth map of each player, using a neural network that – researchers say - is trained on 3D player data extracted from football video games. “We compare with state of the art body pose and depth estimation techniques, and show results on both synthetic ground truth benchmarks, and real YouTube soccer footage,” the researchers added
in their paper.
Essentially we are talking about a system that could, in the future, allow people to view a live football match that is played out, say, on the top of their coffee table. Viewers could then walk around the match and view the action from different angles. It would even allow you to lean in and see the action up close. The effect can be created from just a simple YouTube video.
“One way to create such an experience is to equip the soccer field with many cameras, synchronise the cameras, and then reconstruct the field and players in 3D using multiview geometry techniques,” say the researchers in the paper. “Approaches of that spirit were previously proposed in the literature and even commercialised as Replay’s FreeD, and others. The results of multi-view methods are impressive, however the requirement of physically instrumenting the field with many synchronised cameras limits their generality.”
This approach promises to create the desired effect from a simple 2D YouTube video.
Researchers Konstantinos Rematas, Ira Kemelmacher-Shlizerman, Brian Curless and Steve Seitz have no doubt timed the publication of the paper, in part, to coincide with the 2018 Word Cup. But it’s unlikely we’ll be seeing it in action before Qatar 2022, at the earliest.
The demonstrations shown so far feature footballers jumping around and rarely show the ball (the players’ positions are only estimated and reconstructing the ball “remains a future work”) but it’s a great idea and there’s solid basis for development.
This means football matches could be coming to table top near you soon, assuming other hurdles (such as the ubiquity of “3D viewers” and “augmented reality devices”) can also be overcome.
The paper is being presented at the Computer Vision and Pattern Recognition conference in Salt Lake City, USA.