Three Microsoft project researchers Johannes Kopf, Michael Cohen and Richard Szeliski have developed “First-Person Hyperlapse” an impressive technology. It turns shaky video from the helmet camera or data glasses such as Google Glass into smooth Hyperlapse videos that acts as if the camera man is floating through the air which surprisingly works really well.
In the video above you will first see the original shaky and slow video through the lens of a helmet camera and then the video becomes interesting as the maker of the video timelapses it at 10X speed, smoothing the camera shakes beyond the capabilities of current stabilization techniques.
First-Person Hyperlapse Technical Details:
You can take a first-person video, such as from Go Pro or Google Glass, and speed it up, but unlike regular time-lapse, such as taking every 10th frame, which amplifies every camera shake, this “hyperlapse” system makes very smooth videos. It simulates “time-lapse videos with a smoothly moving camera”.
The system works by creating an approximate 3D structure of the world using a structure from motion algorithm. The next step is finding a camera path through the world that moves the video through a trajectory that mimics the original path while optimizing for smoothness. Finally you render the output by combining multiple input frames to make each output frame.
First-Person Hyperlapse Technical Video: