Let’s paint a picture: It is a beautiful summer afternoon. A cloudless sky hangs over a stadium where two soccer (football) teams are about to play the match of a lifetime. Unable to get tickets to the sold-out event, you and millions of spectators sit on the edge of your couch cushions as the whistle blows to signal the start of the game. Eighty-seven of the scheduled 90 minutes pass with both sides deadlocked in a 1-1 tie. Any mistake could result in victory for the opposition.

But then, as a 50-50 ball is intercepted by one of your team’s strikers, you watch as they make a clean break across the pitch, pivot, and sail a perfectly kicked ball just beyond the goalkeeper’s outstretched fingers to seal the match in the final minute. The crowd goes wild, the announcers shout at the top of their lungs, and you feel the vibrations of millions of feet jumping simultaneously in excitement in living rooms across the world.

Except, you never saw the goal.

As the camera pans to the net, it captures a different camera crew instead, which is perfectly obstructing your view of one of the greatest moments in sports history and leaving those lucky few thousand ticket holders to be the only true spectators of the goal.

Perhaps with similar fears in mind for fans of all sports, a research team from Kaunas University of Technology (Lithuania) developed an end-to-end system that eliminates visual distractions caused by overlapping camera angles. The system detects and classifies objects in images in a single pass, making it ideal for real-time events, such as live sports broadcasts. Using a combination of computer vision, AI, and other technologies, the system analyzes video frames from incoming camera feeds to detect unwanted objects, such as errant cameramen (bad), or the occasional on-field streaker (worse). Each frame becomes an image that the AI element divides into grids to better recognize all objects within them.

A deep learning technique called inpainting is then implemented to systematically remove the unwanted crew from each frame, while simultaneously filling in the negative space with relevant background details. These frames are integrated back with the livestream and sent to your TV — with audiences none the wiser that any type of distraction appeared in the first place.

---

Courtesy of iStock.com/Dmytro Aksonov.

---

The best part: The researchers predict that on TV servers, the algorithm should be quick enough to complete the entire process within a few-second delay while on-air, meaning that not only will it be quicker and easier than scoring on an open net, but it will also prevent additional delays to the standard live broadcast.

The researchers see this as an absolute win for live broadcasts, which they think will have a much more professional feel once the system is implemented.

As for the fans: They will never have to worry about missing another moment of glorious victory from their favorite team.