System That Automatically Converts 2D Videos Into 3D For Soccer Games

A team of researchers from MIT and QCRI aka Qatar Computing Research Institute has come up with a new system that lets you watch a 2D soccer game in 3D on your smartphone is you have a Google Cardboard, your 3D TV or even Oculus Rift. Most people who own 3D television sets these days complain about the lack of content created exclusively for watching in 3D. It turns out that in 3D movies, there are artists for painting the depth in a normal 2D video, something that's not possible in a live soccer game. With the newly developed system the system will automatically convert your 2D videos into 3D and enjoy a more realistic experience.


It's not the first time that researchers have tried to build such a system. However, in most systems that exist till date, the output 3D video contains some odd visual artefacts that ruin the video watching experience. Therefore, while developing a solution, MIT and QSIR researchers thought of targeting a very specific problem, in this case - the soccer sport.

The team was inspired from video games which store detailed 3D maps of a given virtual environment that the user is navigating. When the player makes a move, the map gets adjusted automatically in real time and creates a 2D projection of a 3D scene that's suitable for a particular viewing angle.

They used the famous video game 'FIFA 13' as their reference point. They played the video multiple times and made use of Microsoft's analysis tool called PIX to continuously capture 10s of 1000s of screenshots of the video, while extracting a 3D map for every image. They only selected the best shots that captured the range of viewing angles using a standard algorithm. All this was stored in a database.

Now the system has been developed in such a way that it looks at ten different screenshots in the database for every frame in a 2D soccer game video. It then decomposes those images to find a match between small section of the video to that of the screenshot. By superimposing the depth information from the screenshot to that video and stitching them together, the system creates 3D videos out of 2D content in real time, that too without any visual artefacts.

The MIT and QSIR team is now working on reducing the conversion time in such a way that it becomes fit for broadcast.

What are your thoughts about such a 2D to 3D video conversion system? Share with us in comments below.

Source: #-Link-Snipped-#