1
$\begingroup$

I have two video files with matching background and I like to overlay them. Both are not perfectly stationary, so either one or both need to be stabilized and matched/tracked against each other.

Could someone point me in the right direction of tutorials or keywords to search for?

I've already read up on tracking and overlays, but combining the solutions seems non-trivial. Some involve 3D objects, some are entirely done within the video editor / node view.

Some additional info:

  • its actually top-down drone footage of a racetrack and I want to overlay them to compare racing lines of cars on the track
  • the footage is stable enough that the tracking could be done manually with keyframe animation if automatic tracking is not possible
  • at least scale and rotate is needed for the backgrounds to match
  • ideally I could choose a selected smaller area to render into a final video in a later step
  • being able to use the grease pencil to draw annotations on the video would be a plus
$\endgroup$
3
$\begingroup$

It sounds indeed like a non-trivial solution, that will require you to get proficient with motion tracking and/or compositing.

For a 3D way to do it you would need to track the footage and solve the camera.

The success of the operation would depend entirely on the kind of motion from the cameras, if there is parallax on the scene then you might be able to recreate 3D info. If you track features for which you know the scale in the real world, then use it to set the scale in the 3D reconstruction.

You could then to re-project the image into some geometry, so that you have a reconstruction of the 3d space as fixed points, and a image projected from each of the positions on the camera at a given frame, the UVproject modifier will allow you to use the camera as a projector or you can use a lamp that is parented to the camera to project the image as texture.

You would need to do the same for the other image and manually reposition the camera so that the projected images coincide with those of the other camera.

A non-3D would be to create two un-distorted and stabilized images out of the video. Those new pieces of video can be overlaid using difference mode in a mixRGB node for clarity. One of the videos would need then to be rotated, scaled and possibly distorted so that some the features of the track correspond to those in the other image. Note that the difference in perspective from both images might prevent you from getting a perfect match. You won't be able to invent what the camera did not see.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.