Sphere Registration Failure

[Anonymous]

I recently had the opportunity to register an outdoor scene with spheres.  In remote areas, ReCap cannot self register (nothing to grab onto) and checkboard targets are poorly recognized.  I suspected and told my client that if we hope to get properly registered roadway scenes, spheres are probably the only way to go.  (fragility, cost & portability were my client's objections to spheres).

Since my workaraound for projects that do not register is so cumbersome, we now have 12 spheres.  I used them last week to scan a scene.  I was excited to see three common targets recognized for all the scans (some 50-60 feet away!) and I anticipated seeing a tight set of scans in Project View.

The drum roll turned into a bah-dump-bump.  All the scans are a little off at best and sometimes significantly off.  I 3D Previewed it before indexing and there was no indication of problems. Dang it. 

[ryan.frenz]

It's interesting if 3D preview looked 'better' than the final result (as they should be identical).  We're working on better nav tools so you can really interrogate the result in 3D preview.

I assume you ran the target enhancement (bundle adjustment) at the end.  It's available from the final (single-pane) page at the end of registration, from the fly-out menu on the registration group.  Until this is done you're just seeing cloud-to-cloud results seeded by the target positions.  And based on your description of the scene, that's not ideal.

I almost didn't ask, until I realized that the Help website seems to now be missing this feature.  So sorry about that.

Assuming you did that - what's the overall RMS for the project?  What was the worst scan/target RMS in the registration report.

Ryan

[Anonymous]

I recorded the entire registration process, my investigation and then my AutoCAD workaround, which produces the result ReCap should have produced, but didn't.  I've produced a narrated video showing all this.  I will open a trouble ticket, get a link and upload it to AutoDesk.

As far as RMS data  some where good, some bad.  It seems to me that when target registering, those metrics are meaningless because, for one, I can't do anything about it.  There are three common targets in each set of scans.  If ReCap recognizes them, then a sphere center has been calculated and a 3 Point Align should be perfromed.  No further analysis necessary.  It can give me scan metrics feedback, but I should be accepting the result no matter what.  The result should be accurate if the targets and nothing else comes into play.  If, after all that, a "refinement" pass can tighten it up even more, then great, but the targets should define the registration.

[ryan.frenz]

Thanks, that's helpful.

To be clear - if you want the registration to be based on your targets, you must perform the enhancement at the end.  Otherwise you're just getting cloud-to-cloud registration with the targets used as 'hints' - no different than when you pick red/green/blue points as hints.

The enhancement will also perform a nonlinear optimization that will balance any accumulated error (from measurement uncertainty, extraction error, etc) across the target network.  This allows the target registration to tighten itself based on correlations between scans (similar to what you get for free with cloud-to-cloud on indoor projects).

Ryan

[Anonymous]

I'm trying this scene again, although I find the workflow odd.  I tried to return to the root, unedited scene rcp file and in registration, the enhancement feature is grayed out, so I guess I have to do that after registration but before indexing.  I also found an article showing this process (TargetWorkflow.pdf).  It said, do normal registration without spheres, then go through and pick all the spheres, then run the enhancement.  I find this counterintuitive.  Got spheres? Use them, pick them, done.  Seems like a simple scenario is overly complicated. 

Anyhow, trying it again.

Also, seems like a gotcha setup for anyone expecting straight forward target utilization.  This needs to be better featured.  If it let's me pick spheres at the start of cloud to cloud registration, then it better expect I want to use them and automatically do what is necessary to pull off a target utilized result.  Not that the user has to go though a secondary pass to do the thing you obviously want to do from the start. 

[Anonymous]

Well...I did the normal registration and started adding targets to the completed scan group.  Of 19 scans, of which I could identify more that 4 targets each, scan 11 would only see 2 targets (the third target would work if it didn't jump to the rounded top of a post instead).  For that reason, the target enhancement feature is grayed out.  See attached.

Classic example of why this is a poor method.  18 of 19 scans should be 3D Aligned to three common targets, then scan 11 uses the old 3 point method.  I would say, use the two targets it recognized and then a third pick point (non-target) for a mix-and-match feature.....but that would be asking too much.

So I guess I have to remove Scan 11 to get the target enhancement feature to come online.

What?  Then I enhance, then I try to re-introduce scan 11?

You guys are taking something that should be super simple and convoluting the heck out of it.

[Anonymous]

After removing scan 11, I could enhance the registration.  The RMS seemed to get nice and tight.  I then added scan 11 using the standard 3pt method.

Indexing now..... we'll see......

[Anonymous]

Complete disaster.  I did the enhancement.  Targets identified galore.

This is the worst result yet.  Far worse that what I recorded previously.  See attached.  This is one telephone pole.

[ryan.frenz]

Interesting.  I assume this is after you added Scan 11?  My guess is that scan 11's cloud-to-cloud registration is bad enough that the error gets spread across the network.  You can confirm that quickly by removing 11 & running the enhancement again (no need to re-index).

[ryan.frenz]

Regarding your other feedback - thank you for it, we appreciate and hear all of it.

ReCap Pro's registration tools are designed for cloud-to-cloud registration.  The features relating to target extraction are mainly for independent verification of results (via the target data report) and survey grid alignment.

As you've noticed, support for registering with the targets is quite limited - mostly because only a very small percentage of projects need it.  But your comments are quite valid, and I'll make sure they are considered for future improvements.

Thanks again,

-Ryan

[Anonymous]

By your own statement ReCap does NOT do targeted registration.  Unbelievable!

You should use targets for a 3PT Align (and perhaps greater accuracy if you have 4 common targets etc...).  You don't have to know jack about the scene.  I,2,3 1,2,3 done.  1,2,3 1,2,3 done.  15 minutes later I have a perfectly aligned scene.

Targets are not an enhancement to your auto-alignments.  Well, it is, but shouldn't be the first and sole use of targets in aligning scans.  I think you guys have missed the plot completely.  I have other work to do.  I've taken this as far as I am willing based on what I now know.

Please let me know if you ever start using targets properly.  In the meantime, I will investigate your competitors.

[ryan.frenz · Accepted solution]

I'm sorry that ReCap doesn't work how you expect.  This is valuable feedback.  Thanks for sharing your experience with the community.

Just to close the loop on this thread: the problem in the original post can be solved by using the 'target enhancement' feature - but this does require 3+ shared targets per scan, for the whole network.