We figure out the 2D position of any device on a flat surface using optical flow from the camera and integrated accelerometer data, then display the appropriate image on the device. Instead of scrolling or panning to see a different part of the screen, just move the device across the table. Useful for maps, remote desktop, shared drawing canvas, etc.
Also this is all done client-side in JS, so it works cross platform on mobile devices, laptops, etc. just by browsing to that url and giving it permission to access your camera.