Facebook Inc (NASDAQ:FB) is working on new ways of making sure that 360-degree videos are stable.
As the virtual reality industry gains more steam, 360-degree videos have also become a reality and something that is soon expected to become mainstream. It is still in early development, for now, meaning there are still some problems to be dealt with and one such problem is instability. Facebook has identified that video recording is especially when human hands are involved is a tricky process; particularly because hands are often shaky. Facebook is working on a new way to fix this problem. The argument is that shaky videos may cause nausea to some viewers and such videos are generally not attractive.
Facebook is currently working on a new algorithm that it hopes will make 360-degree videos more stable with the hope that the efforts will improve the experience for the viewer. The algorithm is said to combine the benefits of two-dimensional and three-dimensional algorithms used for stabilizing the videos. “We use 3D analysis only between key frames spaced a few seconds apart in time, and instead of performing a full reconstruction, we only estimate relative rotation, which is an easier problem that can be robustly solved,” said Johannes Kopf, a research scientist from Facebook in a blog post.
The research scientist’s blog post further went on to explain that a 3D algorithm is better because it can determine whether the observed motion comes from the translational or rotational camera motion. For the inner frames (the remaining frames between the key frames), we turn to 2D optimization and directly maximize the smoothness of motion in the video.
We developed a new ‘deformed-rotation’ model, which is similar to a global rotation but allows slight local deformations. We optimize the parameters of this model so it can adapt to handle and undo some degree of translational shake (such as bobbing up and down while walking with the camera), rolling shutter artifacts, lens deformations, and stitching artifacts. In this way, we minimize the effects of these phenomena,” the scientist explained further.
Kopf believes that the new approach will allow the image stabilizer to function better and faster than the regular 2-dimensional and 3D models.