First-person cameras like GoPro and Google Glass have been steadily increasing in popularity over the past several years, and ever since Google Glass was made available to the public, similar forms of attachable first-person cameras have become all the rage. From people recording and uploading mundane experiences like bicycle rides to more significant activities like recording everything you do while going on a month-long road trip through the United States, or even terrifying combat footage captured from a helmet-cam, these small devices provide captivating perspectives to the public that are timeless and fascinating.
A still taken from a GoPro in time-lapse mode simulates the appearance of driving at warp speed.
Another trend gaining popularity is recording videos in time-lapse mode, which is essentially speeding up the footage to about 10x the normal speed. Time-lapse videos are normally accomplished using stationary cameras instead of first-person attachable cameras. Due to the movement of the wearer, the time-lapse footage typically taken by attachable cameras is extremely shaky producing confusing images at such a high speed.
Recently, a team of researchers from Microsoft have developed an algorithm that completely smooths the footage, called “hyperlapse.” There is nothing simple about an algorithm, especially one that corrects an issue that was previously thought to be impossible to achieve.
As an explanation on why attempts to fix this problem haven’t been successful until now, the research team at Microsoft asserts, “At high speed-up rates, simple frame sub-sampling coupled with existing video stabilization methods does not work, because the erratic camera shake present in first-person videos is amplified by the speed-up.”
The following video is a demonstration of ‘hyperlapse’ which corrects the previous problem of shakiness, and in my opinion, to say that it is amazing is a serious understatement:
This video provides a more technical explanation of the system:
For a full explanation, including diagrams, please visit the Microsoft research team’s website.
