First-person videos have become a norm thanks to head-mounted cameras like the Go Pro and Google Glass. It's a great way to capture action; the only problem is this same footage is often so shaky it turns into a nausea-inducing mess.
Now software engineers at Microsoft Research (opens in new tab) have developed an algorithm that converts first-person footage into a smooth hyperlapse no matter how jerky the original recording.
The resulting video removes any errant shakes, head bobs and other stray movements for a much smoother viewing experience.
Microsoft demonstrated the robustness of its algorithm by converting a bike ride into what looks like a flying hyperlapse. Meanwhile, the same process turns a mountain climb into a silky pan across a cliff side.
The hyperlapse method essentially looks at the footage to create a 3D path based on how the camera-wearers were moving in a scene. The algorithm uses this information to trace a smooth path while stitching together multiple images into new frames along this stabilized trajectory.
The footage isn't perfect. From the few clips we can see some obvious artifacts including one rock climber appearing in one frame to miraculously disappear in the second. The way images are stitched together also causes details to suddenly appear in an unnatural way.
However, as head-mounted recording gadgets seem like an inevitable part our wearables future, Microsoft's algorithm could lead to much smoother video first-person videos.
For now Microsoft's hyperlapse tool only exists as a research project, but the group is working hard to release a Windows app. We can only hope it comes out soon.