This is an awesome bit of research developing a method of reconstruction a Hyperlapse from ordinary video shot in the First Person View. A steady cam gets rid of shaking that occurs over a fraction of a second. This algorithm is designed to remove motion over many seconds. It works based on the fact that most things don't move. The ground doesn't move, houses don't move, mountains don't move. It is able look look at these things that don't move and guess how they look even when the camera is turned away. Things that do move, such as walking people, look weird because their is not physically enough information for the computer to figure out how they looked when they were out of frame.

We present a method for converting first-person videos, for example, captured with a helmet camera during activities such as rock climbing or bicycling, into hyperlapse videos: time-lapse videos with a smoothly moving camera.

This video accompanies our SIGGRAPH paper and provides a technical explanation of our system.

Check out our project page for more details: http://research.microsoft.com/hyperlapse