Microsoft Research was manually computing a motion map using heavy image-processing techniques that they then used to decide which frames to use and how to align them. Their work can do things like throw away a series of 'noisy' frames that wouldn't fit the overall scene motion. For instance, if you had a head mounted camera and were riding around on your bike, it could throw away the moments where you look both ways before crossing the street.
Instagram's technology utilizes extra sensor data to make their frame corrections. As far as I can tell from the brief article, the technology doesn't do much besides stabilization. That's not to belittle it, as it's a really neat application demonstrating the benefits of mobile sensor integration. And you might be able to create a similar motion map using the gyroscope data that the folks at Microsoft create through their image processing techniques!