That's nice!
To begin with, you will need about 30 devices
I imagine that all of them are receiving the same unique content by multicast (it mixes pre-made video with real preview from a cam)
This content has some time of timestamp for synchronisation, and devices are "reasonably well" time synchronized
Each one is aware of its position, dimensions and orientation
So each one only displays the piece that corresponds to their relative position window.
The content creation process, being aware of the final subwindow distribution and taking advantage of it, makes a part of the magic.