Microsoft's Research division in Redmond, Washington has published
material detailing a method for turning shaky first-person camera
footage into a stabilized hyperlapse video. Hyperlapsing is a technique
used in multimedia to string together a collection of photographs or
video frames in order to produce a cinematic experience; the methodology
is also more commonly known as a stop-motion time-lapse.
Microsoft researchers, Johannes Kopf, Michael Cohen, and Richard
Szeliski, all noted that first-person video cameras could capture
interesting footage during activities such as 'rock climbing or
bicycling', but can be 'dead boring' to watch at normal speeds.
The three researchers found that simply speeding up the video by a
factor of 10 (also known as subsampling), did not produce the smooth
result they wanted, but instead an extremely shaky video where every
movement was amplified. When the researchers attempted to, instead,
stabilize the video, they found that they simply didn't have enough
information to do so.
The algorithm they've developed uses a 3-step process to produce
exceptionally stable videos that play back in a single fluid camera
sweep. First researchers reconstruct the scene by creating a 3D model of
the world based on similar structures detected within each frame. Next,
they proceed to plan a camera path through the world and create 3D
point-clouds from overlapping frames. Lastly, the dense depth maps are
stitched together, color changes are made, and the video is rendered for
viewing.
The resulting video, seen in the demo below, is quite impressive. As
of now, the team does not have a release date, but mentions that those
who are interested should stay tuned as they 'are working hard on making
[their] hyperlapse algorithm available as a Windows app' - Note that
the word app most likely hints at a Windows 8 application and not legacy
desktop software (sorry Windows 7 users).
Video Link - https://www.youtube.com/watch?v=sA4Za3Hv6ng
0 comments:
Post a Comment