Here's the mirror way of doing it:
There's a link on the home page showing what the uncorrected video looks like (it's a giant video donut).
You get better image quality (e.g. no stitching errors), but lower resolution, because the whole environment has to fit in the same 1080P frame.
http://en.wikipedia.org/wiki/Omnidirectional_camera http://www.cis.upenn.edu/~kostas/omni.html
This is getting very popular in robotics applications because it's cheap and a simple way to get 360 surroundings data. From this, you can localize and / or detect obstacles.