I love my GoPro cameras. One is a constant companion in my backpack. A few years ago, I had an idea to shoot a behind-the-scenes time lapse of my near constant travel (e.g. coming/going to the airport, planes taking off/landing). I put fifteen or so videos in the can but soon gave up on the project.
The reason was simple. They sucked.
Why? The camera was always attached to a moving, unstable object like my backpack or roll-aboard. As a result, the image jerked around too much and, at the time, the GoPro frame was too small to give me enough “slop” to stabilize the footage and still render a final product at 720p. I could have corrected this in post, but the process would have been far too time-consuming to merit the work. So, the frames and footage sit on a hard drive, untouched.
Today, I came across a technology called Hyperlapse from Microsoft that could bring new life to my project. My old footage would likely remain on the shelf because Hyperlapse solves the problem a different way. Have a look.
My original project had a basic requirement shoot from an unstable, moving platform (e.g. the human body, a rolling suitcase) with two basic options, each with their own challenges.
- Shoot Video: This gets me tons of frames, but I would run out of memory and drain the battery more quickly and have a smaller frame with less room for motion stabilization.
- Shoot Time Lapse: This gets me a larger frame, which would mean more room for stabilization in post, but I end up with far fewer frames. While this means I have better battery life and storage capacity, it also results in less pleasing transitions and a “jerkier” final video.
Today, I would simply record video because the magic triangle of frame size, memory capacity and battery life have all dramatically increased. But, one problem would remain; constant, jerking motion in the frame.
This is where Hyperlapse shines.
Hyperlapse analyzes and reverse engineers the motion path of the video. Then, rather than grabbing frames at a set interval (e.g. every 10th frame), it selects the next frame that best lines up to the previous frame, even if that frame is incomplete. If I understand correctly, the software then samples unused frames to complete partial frames and, perhaps, do some inbetweening.
So, unlike traditional motion stabilization which is inherently destructive, Hyperlapse is constructive. This results in considerably smoother transitions in the time lapse itself and a much cleaner frame without stabilization and cropping in post.
The only negative I see is the presence of apparent artifact frames in the test video. Hyperlapse is still in development, so a fix is probably planned. If not, it would be a simple process to remove artifact frames from the final render.
All in all, pretty impressive. The results speak for themselves.