trouble refining the registration with target spheres

Discuss FARO SCENE software here.
Post Reply
skybornevisions
I have made 10-20 posts
I have made 10-20 posts
Posts: 11
Joined: Fri Jul 10, 2020 5:13 pm
3
Full Name: Steve
Company Details: KSC
Company Position Title: CAD
Country: USA
Linkedin Profile: No
Has thanked: 2 times
Been thanked: 1 time

trouble refining the registration with target spheres

Post by skybornevisions »

I'm pretty much a newbie--trying to learn on the job for the past 6 months using a Faro Focus 3D X330 and Scene 2019.2 to put together point clouds for the exteriors of buildings to use as as-builts for a construction company. Most of my training has come from a Faro salesperson and a patient tech support person; so I mostly operate in a constant state of confusion; so I apologize in advance for asking silly questions.

I've noticed that after an initial registration using target spheres (I use 6 target spheres with a 3 target sphere over-lap), my registration report is "all in the green"; and the point clouds themselves seem nicely aligned with each other. However, if I start refining the registration by setting anti correspondence of target tensions with the highest distance errors, I've noticed at times that even while the distance errors trend downward (as desired); sometimes one or two scans seem to get way out of alignment. The first time I noticed this, one of my scans was off by over 2" from the rest; but by the time I finished getting the lowest distance error; it all seemed to come back together.

However, with my latest project, I noticed two of the scans (out of 26) would quickly go out of plane by around 3/8" after the second anti registration, and never improved by the time I went from a distance error of 0.357" to 0.044". Unfortunately due to the graphics in Scene, this wasn't obvious until I tried working with the point cloud in AutoCAD--which I use to take measurements with. I decided to delete the two offending scans, since I seemed to have enough overlap anyway. So far that seems to have helped. However, this process is very time consuming (the export process to RCP takes awhile); plus it's hard to check every nook and cranny of the point cloud (in a timely manner) to see if everything is truly aligning correctly.

Does anyone have any tips they could offer? Perhaps I should just be "happy" with initial registration reports; instead of trying to chase a perfectionist report that obviously doesn't correlate directly to tighter registrations between scans...

I've seen a couple similar threads from 4 years ago; but none of them seemed to offer a specific solution nor address the problem that I'm facing. I've seen some say they use target spheres initially followed by cloud to cloud.

I noticed that yankoch posted here: https://laserscanningforum.com/forum/vi ... 570#p51570
yankoch wrote:Then, when the target based registration is ok, we just drag the target based scans, keeping their position, into C2C folders, and we define them as the reference scans. Then we do C2C using these reference scans.
Perhaps this might help me, but I have no idea how to drag scans into C2C folders. I've got lots of experience using C2C in the past; but it seems to over-write any prior effort with registration target spheres first. Can anyone direct me to a tutorial on how to do this process?

I'm hesitant to disable inclinometers for various scans (though it might be better than simply deleting them; assuming this helps); as I understand that by keeping inclinometer enabled during registration helps accuracy: a function of the inverse square law of 3D fields. Knowing what's truly vertical is very important for what I do as I measure how far out of plumb walls are.

Appreciate any help on this matter...

Thanks,

Steve
User avatar
jcoco3
Global Moderator
Global Moderator
Posts: 1724
Joined: Sun Mar 04, 2012 5:43 pm
12
Full Name: Jonathan Coco
Company Details: Consultant
Company Position Title: Owner
Country: USA
Linkedin Profile: No
Has thanked: 70 times
Been thanked: 157 times

Re: trouble refining the registration with target spheres

Post by jcoco3 »

Hi Steve,

I just read your post and right of the bat if you are having to chase registration issues by setting a bunch of anti-correspondences then you probably have something else going on. Sounds like you got taught a simple 3 sphere between two scan pairwise matching approach that can lead to some issues once you project gets larger or has lots of overlapying loops that don't share enough targets. If you sphere constellations ate very similar then it could be your smoking gun for the mixed up correspondences. Really it could be something else entirely though. Posting a few pictures to start with this type of problem usually goes a long way.
so I mostly operate in a constant state of confusion; so I apologize in advance for asking silly questions.
What's the fun in knowing everything prior to the journey anyway? Its ok, we are all confused most of the time.
User avatar
landmeterbeuckx
V.I.P Member
V.I.P Member
Posts: 1615
Joined: Tue May 01, 2012 5:19 pm
11
Full Name: Lieven Beuckx
Company Details: Studiebureau Beuckx
Company Position Title: Owner
Country: Belgium
Linkedin Profile: Yes
Has thanked: 183 times
Been thanked: 548 times

Re: trouble refining the registration with target spheres

Post by landmeterbeuckx »

It is a sphere problem like Jonathan said, a lack of amount of spheres. The 3 number only works when setting them with progress in mind. At some point this will work against you.

Use at least 10 so you overlap more spheres. Also change the height, do not place them all on ground surface but one on ground, one on a pole, one at 2 meters,...

This will normally solve your problem.
LSBbvba
Surveying services - 3D Laserscanning
Tel : +32477753126
www.lsbbvba.be
[email protected]
fobos8
V.I.P Member
V.I.P Member
Posts: 246
Joined: Sat Jul 07, 2018 9:19 pm
5
Full Name: Andrew
Company Details: NDC Surveys
Company Position Title: Surveyor
Country: Uk
Linkedin Profile: No
Has thanked: 56 times
Been thanked: 22 times

Re: trouble refining the registration with target spheres

Post by fobos8 »

Hi Steve
What resolution setting did you use?
What distance from the scanner were the spheres?
What size spheres did you use?

Some spheres may apear to have poor target tensions but if your resolution was low or the sphere a long distance away that will happen..

As an alternative approach, it may be worth trying c2c to see if you get good results. You can get Scene to calc target stats for the spheres for c2c as a check on your registrations.

I often get better registration in Scene with c2c, particularly when the spread of spheres is poor.
Regards Andrew
skybornevisions
I have made 10-20 posts
I have made 10-20 posts
Posts: 11
Joined: Fri Jul 10, 2020 5:13 pm
3
Full Name: Steve
Company Details: KSC
Company Position Title: CAD
Country: USA
Linkedin Profile: No
Has thanked: 2 times
Been thanked: 1 time

Re: trouble refining the registration with target spheres

Post by skybornevisions »

jcoco3 wrote: Sat Jul 11, 2020 4:45 am Hi Steve,

I just read your post and right of the bat if you are having to chase registration issues by setting a bunch of anti-correspondences then you probably have something else going on. Sounds like you got taught a simple 3 sphere between two scan pairwise matching approach that can lead to some issues once you project gets larger or has lots of overlapping loops that don't share enough targets. If you sphere constellations ate very similar then it could be your smoking gun for the mixed up correspondences. Really it could be something else entirely though. Posting a few pictures to start with this type of problem usually goes a long way.
Here's some screenshots: The first set are screenshots from a more complex building (Nolan Idea which consisted of a two story building made of two concentric arcs) where I first started using target spheres for registration, and where I noticed some mis-alignment between scans even as my distance errors went down. I had been setting up target spheres on this project and previous ones, but never utilized them correctly; as I did all my registrations c2c. On this project "Nolan Idea Bldg", I couldn't get good tolerance reports after spending all day interpolating c2c settings (keeping several pages of notes of what settings gave me more accuracy)... and that's when I got the attention of a Faro tech rep [who's been unresponsive so far this time around--probably fatigued at being my personal unpaid tutor] who told me I wasn't even utilizing my target spheres (at the time I thought their only benefit was to give a general alignment of scans to help c2c out)... and taught me the benefit of of setting anti correspondences... So, in this case, I felt "forced" to use target spheres for registration as c2c just wasn't able to get everything "green". Everything seemed to correctly align in the end with this project after I set enough anti-correspondences to get a very low distance error... The reason I'm pretty sure, is that I'm also tasked with extracting dimensions and doing a layout overlay on point cloud; providing notes where shims or furr-outs may need to be provided... which requires close inspection of the point cloud... Plus, the crew in the field confirmed that the places I had marked as being out of plumb were indeed out of plumb... Also this was the first project (after about 8 less complex ones) where I didn't have to re-make any panels due to bad dimensions--actually first project where no panels had to be re-made for any reason which is unusual; especially where most of the surfaces we covered were curved. So for what ever reason (beginners luck) I have had some success using the workflow on a more complex building with more scans, than the project I'm now having trouble with: Strickland.
nolan-high-idea-scene.jpg
nolan-idea-panel-perspective.jpg
Here's some screen shots of the Strickland project. A quick note on this... I discovered that by disabling the inclinometers on "scan008" and "scan015", the scans would align better in the "problem areas" (Fig. 1 and Fig. 2); but then go out of alignment elsewhere--though not as bad as before. I've also thought of using the point cloud without the two scans (as I seem to have enough coverage with the surrounding scans); but this is a "last resort" solution for me.
strickland-north-scene.jpg
strickland-north-west-iso.jpg
strickland-north-west-persp.jpg
FIGURE1.jpg
strickland-fig2.jpg
I'm including a perspective view to kind of show the placement of scans and target spheres. As you can see, I've tried adding some vertical displacement (though far from ideal I'll admit); from an existing neighboring parapet to mounds of dirt at ground level... and use of a camera tripod for a couple spheres (that my be contributing to the problem due to being too light weight for being vibration resistant in a light breeze). Oh, and the pickup truck was moved prior to scanning "scan015".

Hope these screenshots are helpful...

Thanks for your response,

Steve
You do not have the required permissions to view the files attached to this post.
skybornevisions
I have made 10-20 posts
I have made 10-20 posts
Posts: 11
Joined: Fri Jul 10, 2020 5:13 pm
3
Full Name: Steve
Company Details: KSC
Company Position Title: CAD
Country: USA
Linkedin Profile: No
Has thanked: 2 times
Been thanked: 1 time

Re: trouble refining the registration with target spheres

Post by skybornevisions »

landmeterbeuckx wrote: Sat Jul 11, 2020 6:35 am It is a sphere problem like Jonathan said, a lack of amount of spheres. The 3 number only works when setting them with progress in mind. At some point this will work against you.

Use at least 10 so you overlap more spheres. Also change the height, do not place them all on ground surface but one on ground, one on a pole, one at 2 meters,...

This will normally solve your problem.
What would you consider the minimum vertical distance between spheres? I did try to place my spheres on different horizontal planes from each other; though places like the roof top (parapet) I didn't have much option other than a camera tripod I brought along, and magnetically attaching another sphere to coping to keep some distance from another mini-tripod... however that variation was only a matter a couple inches. On the ground, I'd try stacking on various mounds of dirt that varied in height, or a nearby fence at various heights.

So, you're suggesting that at any one time, one should have at least 10 target spheres within range of a scanner, with a 5 sphere overlap?

Thanks.
User avatar
landmeterbeuckx
V.I.P Member
V.I.P Member
Posts: 1615
Joined: Tue May 01, 2012 5:19 pm
11
Full Name: Lieven Beuckx
Company Details: Studiebureau Beuckx
Company Position Title: Owner
Country: Belgium
Linkedin Profile: Yes
Has thanked: 183 times
Been thanked: 548 times

Re: trouble refining the registration with target spheres

Post by landmeterbeuckx »

skybornevisions wrote: Sat Jul 11, 2020 6:18 pm
landmeterbeuckx wrote: Sat Jul 11, 2020 6:35 am It is a sphere problem like Jonathan said, a lack of amount of spheres. The 3 number only works when setting them with progress in mind. At some point this will work against you.

Use at least 10 so you overlap more spheres. Also change the height, do not place them all on ground surface but one on ground, one on a pole, one at 2 meters,...

This will normally solve your problem.
What would you consider the minimum vertical distance between spheres? I did try to place my spheres on different horizontal planes from each other; though places like the roof top (parapet) I didn't have much option other than a camera tripod I brought along, and magnetically attaching another sphere to coping to keep some distance from another mini-tripod... however that variation was only a matter a couple inches. On the ground, I'd try stacking on various mounds of dirt that varied in height, or a nearby fence at various heights.

So, you're suggesting that at any one time, one should have at least 10 target spheres within range of a scanner, with a 5 sphere overlap?

Thanks.
Just change the height, it doesn't have to be specific but variation.

You don't have to see 10 spheres at a time but plan in advance where you place them. I'll always try to place the scanner in a constellation of 4.
LSBbvba
Surveying services - 3D Laserscanning
Tel : +32477753126
www.lsbbvba.be
[email protected]
skybornevisions
I have made 10-20 posts
I have made 10-20 posts
Posts: 11
Joined: Fri Jul 10, 2020 5:13 pm
3
Full Name: Steve
Company Details: KSC
Company Position Title: CAD
Country: USA
Linkedin Profile: No
Has thanked: 2 times
Been thanked: 1 time

Re: trouble refining the registration with target spheres

Post by skybornevisions »

fobos8 wrote: Sat Jul 11, 2020 8:12 am Hi Steve
What resolution setting did you use?
What distance from the scanner were the spheres?
What size spheres did you use?

Some spheres may apear to have poor target tensions but if your resolution was low or the sphere a long distance away that will happen..

As an alternative approach, it may be worth trying c2c to see if you get good results. You can get Scene to calc target stats for the spheres for c2c as a check on your registrations.

I often get better registration in Scene with c2c, particularly when the spread of spheres is poor.
Regards Andrew
Hi Andrew,

I used a resolution of 1/4 (43.7 Million points), quality 3x.
I kept the six spheres less than 50 feet away from the scanner for each scan.
The target spheres are 150mm (6") diameter.

The main reason I started using target spheres for registration (besides not realizing I wasn't using them to begin with); was because I could NOT get good results with c2c; even after spending more than a day trying to interpolate results with the combination of settings. With this current project I can't get very good results with c2c either; though I haven't spent much time playing with the subsampling and max search distance combination settings on this particular one.

Can you explain a little more about what you mean by target stats as a check for c2c? (and how one would do that)?

Thanks,

Steve
skybornevisions
I have made 10-20 posts
I have made 10-20 posts
Posts: 11
Joined: Fri Jul 10, 2020 5:13 pm
3
Full Name: Steve
Company Details: KSC
Company Position Title: CAD
Country: USA
Linkedin Profile: No
Has thanked: 2 times
Been thanked: 1 time

Re: trouble refining the registration with target spheres

Post by skybornevisions »

landmeterbeuckx wrote: Sat Jul 11, 2020 6:30 pm
skybornevisions wrote: Sat Jul 11, 2020 6:18 pm
landmeterbeuckx wrote: Sat Jul 11, 2020 6:35 am It is a sphere problem like Jonathan said, a lack of amount of spheres. The 3 number only works when setting them with progress in mind. At some point this will work against you.

Use at least 10 so you overlap more spheres. Also change the height, do not place them all on ground surface but one on ground, one on a pole, one at 2 meters,...

This will normally solve your problem.
What would you consider the minimum vertical distance between spheres? I did try to place my spheres on different horizontal planes from each other; though places like the roof top (parapet) I didn't have much option other than a camera tripod I brought along, and magnetically attaching another sphere to coping to keep some distance from another mini-tripod... however that variation was only a matter a couple inches. On the ground, I'd try stacking on various mounds of dirt that varied in height, or a nearby fence at various heights.

So, you're suggesting that at any one time, one should have at least 10 target spheres within range of a scanner, with a 5 sphere overlap?

Thanks.
Just change the height, it doesn't have to be specific but variation.

You don't have to see 10 spheres at a time but plan in advance where you place them. I'll always try to place the scanner in a constellation of 4.
Okay, I think I understand... I try to keep all six of mine within a 50 foot radius of the scan, and leap-frog as a I go... keeping at least 3 overlapping within range of two scans.
MajorDomo
V.I.P Member
V.I.P Member
Posts: 205
Joined: Mon Feb 25, 2019 4:42 am
5
Full Name: Major Domo
Company Details: VeritasVfx
Company Position Title: ceo
Country: Portugal
Linkedin Profile: No
Has thanked: 17 times
Been thanked: 53 times

Re: trouble refining the registration with target spheres

Post by MajorDomo »

Also, is your scanner calibrated?

X330 seem to suffer from decreasing angular accuracy as time goes by, leading to scans to be in essence skewed.
One way to detect this is by placing all your spheres around you, and scan, lift the whole scanner, and rotate by 45 degrees, repeat a scan, and repeat the process a few more times, if you have a hardware/calibration issue you should see some of the scans out of shape pretty quickly.
Post Reply

Return to “FARO SCENE”