I had a question regarding merging components and how it differs from using the align images with respect to the relative positions of the cameras. I’ll set the stage if you will:
I have a set of images that cover about half a mile of a street. This has been aligned into a single component.
I have a set of LIDAR that cover the same stretch of street, also a single component.
When I merge them together using control points and merge components I discover that the further you move away from the center of the scans the worse the texture - geometry alignment is.
My guess is that there is much more accumulated error in the photogrammetry alignment than there is in the LIDAR alignment.
I suppose that when using Merge Components the images’ positions and orientations aren’t corrected in any way and a “best-fit” scenario is being performed. I imagine that if you were to hit Align Images (but locking the LIDAR in place) you may end up with a much more accurate alignment between the images and the LIDAR but I’m worried that due to the fact that the LIDAR has no color (only intensity) and that there are >20,000 images it will take quite some time for the alignment to complete, if at all.
Is there a streamlined method for aligning two such sets but only doing a minor overall error-adjustment on a photogrammetry set without it having to take several days? What combination of Absolute Pose settings or Relative Position Uncertainty or any others would help here? Or do I totally misunderstand what both Align and Merge do?
Image alignment or image registration is a process in which RealityCapture calculates camera positions, orientations and internal camera states, like camera zooming, for every input image. Once images are registered, you will be able to recognize, how you moved with your camera at the time when you were taking the input images. Before alignment you can change the settings and after alignment you can check the results.
A component is a group of images which are aligned together. Sometimes you can get more components, for example when your photo collection contains more than one disjoint physical object, object texture is weak, there are not enough images or the change in perspective is too high, overlap among images is small and RealityCapture could not pair the images automatically.
There are more options how to connect a scene which was split into more components.
Run alignment again - RealityCapture will first use special algorithms designed for merging components.
Use Merge Components tool - merge already created components. When using this function, no new images are added to the existing components. You can find this tool in the ALIGNMENT tab.
Control Points - with control points you can manually create a link between two or among more images. Provided additional links, RealityCapture can estimate transformations among images or components.
Ground Control Points - geo-referencing of every component to a common space.
Adding More Images - this is one of the faster options.
It is better to use new alignment in your case (after adding control points). Because the image features are detected, it will be quicker than first alignment. Also, the placed control points help to merge these components quicker/better. Also, it can help, when your datasets are georeferenced.