Time-lapse videos don't have to make you sick, and Microsoft researchers have the algorithms to prove it.
Yes, the lure of first-person video is undeniable: Just strap a GoPro to your head or put on Google Glass, and you're set to take your viewers on a wild (or perhaps mundane) ride. If you've watched any of the resulting video, however, you may have noticed that much of it is too long, boring or nausea-inducing.
SEE ALSO: Hold Your Breath as Extreme Paraglider Swoops Between Buildings
Some try to solve this problem by time-compressing the footage, which usually just exacerbates the nausea problem; every bump and camera jerk is sped up and accentuated. Watching the video makes you feel like you're a passenger in a rickety buggy that's careening down a bumpy, winding road at 100 mph. Watching the video makes you feel like you're a passenger in a rickety buggy that's careening down a bumpy, winding road at 100 mph.
Enter the "Hyperlapse" videos created by Microsoft Research. They go just as fast as most time-lapse first-person videos, but instead of jumpy, bumpy imagery, the videos glide through the scenery as if on high-speed rails. Yet what looks easy on screen takes an incredible amount of science and technology on the back end.
hyperlapsepathplan
IMAGE: MICROSOFT RESEARCH
The Hyperlapse technology starts by analyzing the entire video and creating an approximate 3D model of the world. It then builds a camera path through that world that, while mimicking the original path, optimizes smoothness and video quality.
The algorithm looks for dead frames (spots where the camera is not moving forward) and wild deviations from a path forward. It also seeks to readjust the camera, so the viewpoint is not married to your wildly bobbing head but rather looks around in a more casual and controlled manner.
In order to maintain this smooth look, Hyperlapse has to play some tricks with the video. In order to maintain this smooth look, Hyperlapse has to play some tricks with the video. Since the optimal path through the video isn't always the same as where your head is pointed, there will be gaps in the video or landscape coverage (it can't show what you didn't film). Hyperlapse closes those gaps by sampling high-quality imagery from frames adjacent to the ones it plans on using in the final video. It then essentially grafts the necessary image portions into those frames and even blends the stitches and balances the color to create frames throughout.
You can see the results of Microsoft Research programming jujitsu in the video at the top. If you like what you see as a Windows user, you're in luck: The next stop for Hyperlapse video technology is Microsoft Windows. There's no timetable for that development, though.
Yes, the lure of first-person video is undeniable: Just strap a GoPro to your head or put on Google Glass, and you're set to take your viewers on a wild (or perhaps mundane) ride. If you've watched any of the resulting video, however, you may have noticed that much of it is too long, boring or nausea-inducing.
SEE ALSO: Hold Your Breath as Extreme Paraglider Swoops Between Buildings
Some try to solve this problem by time-compressing the footage, which usually just exacerbates the nausea problem; every bump and camera jerk is sped up and accentuated. Watching the video makes you feel like you're a passenger in a rickety buggy that's careening down a bumpy, winding road at 100 mph. Watching the video makes you feel like you're a passenger in a rickety buggy that's careening down a bumpy, winding road at 100 mph.
Enter the "Hyperlapse" videos created by Microsoft Research. They go just as fast as most time-lapse first-person videos, but instead of jumpy, bumpy imagery, the videos glide through the scenery as if on high-speed rails. Yet what looks easy on screen takes an incredible amount of science and technology on the back end.
hyperlapsepathplan
IMAGE: MICROSOFT RESEARCH
The Hyperlapse technology starts by analyzing the entire video and creating an approximate 3D model of the world. It then builds a camera path through that world that, while mimicking the original path, optimizes smoothness and video quality.
The algorithm looks for dead frames (spots where the camera is not moving forward) and wild deviations from a path forward. It also seeks to readjust the camera, so the viewpoint is not married to your wildly bobbing head but rather looks around in a more casual and controlled manner.
In order to maintain this smooth look, Hyperlapse has to play some tricks with the video. In order to maintain this smooth look, Hyperlapse has to play some tricks with the video. Since the optimal path through the video isn't always the same as where your head is pointed, there will be gaps in the video or landscape coverage (it can't show what you didn't film). Hyperlapse closes those gaps by sampling high-quality imagery from frames adjacent to the ones it plans on using in the final video. It then essentially grafts the necessary image portions into those frames and even blends the stitches and balances the color to create frames throughout.
You can see the results of Microsoft Research programming jujitsu in the video at the top. If you like what you see as a Windows user, you're in luck: The next stop for Hyperlapse video technology is Microsoft Windows. There's no timetable for that development, though.
