Been having some inconsistent results with automatic alignment of laser scans to photos.
Wondering if there are any suggestions for getting those to align without relying on CPs. I’ve read through a number of posts here about the topic, but couldn’t find anything specific.
One of my recent projects involved a bunch of scans with a Faro scanner and bunch of photos.
Laser process is as follows:
- All the scans are combined in Scene and exported as an ordered e57.
- Import the e57 into RC (settings below)
- Run alignment on the scans. Everything aligns OK. All the .lsp files seem fine and the “cubes” are in appropriate positions. lsp files looks something like this:
- Then, import the photos
- Align again
This usually results in 2 components. 1 for lasers and 1 for photos. Then I add CPs to combine comps.
Feels like I’m missing something. In the official RC laser+photo alignment video everything just works so smoothly
So, does anyone do anything special to ensure good laser-photo auto alignment?
you need to think of the 6 images per scan (RCs way of converting them) as normal photogrammetry images and connect them in small increments with the rest of the images…
That doesn’t qualify as special at all ! That’s just common sense.
I am a little surprised by how inconsistent the alignment is. Within the same set of scans/photos, sometimes it aligns, other not. Sometimes it finds lots of matches between photos and scan photos, sometimes none. There seems to be something inherently non-deterministic about the process.
I wasn’t sure if you were aware of the way RC treats the scans. I’m not surprized (since you know your stuff), but there were so many questions lately that I’m not on top of everyones personal level of knowledge anymore. I hope you don’t take it personal.
Have you considered that it may be due to the difficult objects you are modeling? For example in your screenshot there is the is the grey surface on the left with a vague reflection - that alone is a nightmare for the algorythms. Then how many blue cylinders with yellow railings are there in the whole scene? We can analyze the content, but RC not (yet?), so it can only look at all those baffling numbers of possible matches that make no sense. So I guess in many cases RC just determines that there is no viable solution and drops scan positions and images from the component.
So to answer your original qestion: No, I would not know of anything to do in your case that I would qualify as special. If you catch my drift…
The only idea is basically what I suggested a few days ago, as in break it down into smaller chunks with less potential for confusing identical or very similar parts of the model.
Oh, and just out of curiosity: why do you import the scans with “Georeferenced” set to false? In my interpretation that would mean that RC drops all absolute coordinates…
Those difficult objects have been a pain for me for a while. That’s what the laser scanner is for I can see how RC would just “give up” after a while of looking at those.
I have been making everything into smaller components and then combining them together and with the laser scans via lots of CPs.
Regarding Georeferenced = False, I just leave that by default. I have GCPs in this project for more precise GPS anyway. Plus I’ve had some issues with setting it to true before, where the scans fail to align.
A number of posts here suggest setting Georef = True or setting the scan rig priors to Locked. I haven’t noticed any practical differences in results by toggling those nor seen explicit explanations as to why those should be set so. Any ideas?
I know and I feel with you every time I see thoes pumps and containers!
In my opinion setting Georef = True means RC won’t change any coordinate of the scan. Georeferencing is a major benefit of scans so I wonder why it’s set to False by standard. Could there have been some interference with GPS data in your case? Or somethimes things don’t work for completely different reasons.
Anyway, if set to false my common sense would tell me that RC can change the orientation and scale of the scans to be able to better combine it with the images. There could also be cases where you have some geo-refed scans and some that aren’t. Maybe RC decided that even if you told it that it CAN change the orientation of your scans it will keep them because it seems ok and there isn’t anything better? Ah, did you delete ALL other components when playing around with this? Because if not, then any existing component might have influenced the others…
