• 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5

Image Alignment?

Offline
#1 suncloud
I was wondering if a feature like this could be added to LRT. A way to "align" the frames/images to correct extremely small movements of camera/tripod during a windy day, for example. It would very similar to "Align Layers" in Photoshop . . . where images are shifted to "line up" (pixel positioning) . . . would this be possible, feasible or a valuable addon ??? Am loving your software and "vision."
Offline
#2 Gunther
Currently I have no idea on how to do this, maybe with an external plugin, but I didn't find one that was suited yet.
In any case this could only happen on the exported intermediary files (not the raw files).
Imho it's easier and faster to use a stabilization feature in a video editing program (most of them offer a feature like Warp Stabilizer etc.) when producing the final video.
Subscribe to: LRTimelapse Newsletter, Youtube Channel, Instagram, Facebook.
Offline
#3 suncloud
Yes, image stabilizer is a good solution . . . thank you!
Offline
#4 zzyzx
I did a deep-dive on this problem a while back, using OpenCV and a bunch of Python to go through a sequence of stills and align them. I went through three or four reprojection algorithms in OpenCV, which is state-of-the art for computer vision. I tried matching neighboring images to each other and also matching them to a hand-picked reference image to avoid drift over time. I tried narrowing the field-of-view down to important, hand-picked features. I messed with tools like contrast to improve feature detection.

The results were disappointing. It turns out that computer vision algorithms are good at aligning images with medium to large spatial offsets, ranging from ~3 pixels and up. But at the pixel level they are inconsistent. (And they can't handle large time offsets when the lighting changes.) So while my script could fix big problems like when I bumped the camera during shooting, the overall result was jittery and worse than the original uncorrected video. And that was with manual intervention along the way. Automating this would make a mess of your time-lapse.

That's a long way to say that Gunther is right on this one: try a video post-processor to do the stabilization. (Though depending on how bad your input sequence is, you would see some lateral drifting in the stabilized output. That's what I was trying to avoid.)

...also check out: