Cloud Registration Case Study - Cyclone V9.0.2

3DVEM, Leica Cyclone Register 360, FARO Scene, LFM Suite, Autodesk ReCAP, Riegl RiScan Pro, Trimble RealWorks, Etc.
Post Reply
Jamesrye
V.I.P Member
V.I.P Member
Posts: 643
Joined: Mon Aug 11, 2008 4:13 pm
15
Full Name: James Rye
Company Details: Merrett Survey Partnership
Company Position Title: Spatial Analyst
Has thanked: 28 times
Been thanked: 69 times

Cloud Registration Case Study - Cyclone V9.0.2

Post by Jamesrye »

I am using Cyclone V 9.0.2 and I thought I'd share with the forum my evaluation of the new pair-wise registration that the program can use on data import.

Firstly, I recently went and scanned an unstable building (just 33 scans outside and inside) where I used targets to register the data - the worst error was 3mm and I worked in a closed loop (i.e. I started and ended on the same 3 targets).

In order to test the accuracy of the cloud to cloud alignment in Cyclone, I re-imported the project and used the new auto-align feature, which registered some groups of scans and on some, it couldn't identify overlaps.

I used the new 2D alignment tool (which I think is great! - much easier and faster than picking 3 points) to join the groups together and then bring in the un-grouped scans. Then I checked the data by slicing it and coloring the scans. I found some mis-aligned ones, but this was down to user error. I had aligned the scans in 2D from the plan view, but not checked the elevation view.

Then to check the accuracy, I registered onto the control for the registration that used targets, I was quite impressed with the results as the mean absolute error was 8mm. See Pairwise_Reg.jpg below.
Pairwise_Reg.jpg
But - I wanted better alignment, so I used the auto-add cloud constraints tool so that it wasn't just a pairwise registration and then I registered that to the targets and the mean absolute error dropped to just 3mm - see image 'Multiple_C2C_to_targets.jpg".
Multiple_C2C_to_targets.jpg
In general, I found the process to be quick and effective. It would be really nice to be able to slice in the 2D alignment tool and also to increase the point size (I have passed this suggestion on to Leica technical support). It's absolutely critical to check your cloud to cloud by taking slices because you can get wonderful residuals on the graph and think everything is OK - only to be surprised that the data is misaligned. I do think that going forward I might use C2C on some projects, but I will always want some targets to start and end on as an independent check. For really high precision work I would always use targets and augment my registration with C2C where appropriate.
You do not have the required permissions to view the files attached to this post.
User avatar
stevenramsey
Global Moderator
Global Moderator
Posts: 1937
Joined: Sun Aug 12, 2007 9:22 pm
16
Full Name: Steven Ramsey
Company Details: 4DMax
Company Position Title: Technical Specialist Scanning
Country: UK
Skype Name: steven.ramsey
Linkedin Profile: Yes
Location: London
Has thanked: 30 times
Been thanked: 72 times
Contact:

Re: Cloud Registration Case Study - Cyclone V9.0.2

Post by stevenramsey »

Jamesrye wrote: In general, I found the process to be quick and effective. It would be really nice to be able to slice in the 2D alignment tool and also to increase the point size (I have passed this suggestion on to Leica technical support). It's absolutely critical to check your cloud to cloud by taking slices because you can get wonderful residuals on the graph and think everything is OK - only to be surprised that the data is misaligned. I do think that going forward I might use C2C on some projects, but I will always want some targets to start and end on as an independent check. For really high precision work I would always use targets and augment my registration with C2C where appropriate.
James thanks for the feed back, great to see such a good write up.

Are you asking for a slicing tool in the 2D visual alingment or in view interim results as 2D slicing is now in Cyclone register as well as point size.
Steven Ramsey

Home [email protected]
Work [email protected]
Mobile +44 7766 310 915
Jamesrye
V.I.P Member
V.I.P Member
Posts: 643
Joined: Mon Aug 11, 2008 4:13 pm
15
Full Name: James Rye
Company Details: Merrett Survey Partnership
Company Position Title: Spatial Analyst
Has thanked: 28 times
Been thanked: 69 times

Re: Cloud Registration Case Study - Cyclone V9.0.2

Post by Jamesrye »

Steven,

I was after a slicing tool in the 2D visual alignment along with being able to adjust the point size. I am using Cyclone V9.0.2 - which I think is the latest version?

Whilst I'm singing Leica's praises - I should also mention that the unify procedure seems to be much faster now as well.
User avatar
stevenramsey
Global Moderator
Global Moderator
Posts: 1937
Joined: Sun Aug 12, 2007 9:22 pm
16
Full Name: Steven Ramsey
Company Details: 4DMax
Company Position Title: Technical Specialist Scanning
Country: UK
Skype Name: steven.ramsey
Linkedin Profile: Yes
Location: London
Has thanked: 30 times
Been thanked: 72 times
Contact:

Re: Cloud Registration Case Study - Cyclone V9.0.2

Post by stevenramsey »

Could you drop me a email with your thoughts on why you would need a 2D slicing tool in visual alignment.
Steven Ramsey

Home [email protected]
Work [email protected]
Mobile +44 7766 310 915
Patrick_C
V.I.P Member
V.I.P Member
Posts: 165
Joined: Mon Jun 06, 2011 2:57 pm
12
Full Name: Patrick Collins
Company Details: Michael Gallie and Partners
Company Position Title: Consultant
Country: UK
Linkedin Profile: Yes

Re: Cloud Registration Case Study - Cyclone V9.0.2

Post by Patrick_C »

This is exactly the sort of discussion we need to see on the forum. Registration methodology and terminology regarding "means", "maximums", "absolutes" etc.

Whilst the principals are good we would not accept a 11mm "bust" on one of the targets. This would imply a local registration error which needs to be fixed.

"mean absolute error" does not negate local errors and whilst the overall project may be good (and therefore may be "fit-for-purpose"). If your original registration had a maximum error of 3mm would you accept a 11mm error on this registration? On the basis it has already registered to 3mm, thereby confirming your target control accuracy, this would imply there is probably an incorrect constraint created by the "auto-align feature".

Whilst we would generally concur with your observations regarding "auto-align" and of the "2D alignment tool" one of the weaknesses currently in the new version is the ability to add incorrect constraints as part of both processes (although more likely in the 2D alignment tool), something not previously possible.

This means that the overall numbers can seem quite good but local registration is incorrect.

Patrick
Jamesrye
V.I.P Member
V.I.P Member
Posts: 643
Joined: Mon Aug 11, 2008 4:13 pm
15
Full Name: James Rye
Company Details: Merrett Survey Partnership
Company Position Title: Spatial Analyst
Has thanked: 28 times
Been thanked: 69 times

Re: Cloud Registration Case Study - Cyclone V9.0.2

Post by Jamesrye »

Patrick,

I agree with you that I'd like to see more discussion regarding registration methodology and terminology. Certainly the registration is not "as good" or "as tight" as the one that used targets. There is always a push to be faster and cheaper - but it's important to be "right" as well.

However, for this project, we had to produce a 1:100 topo and a 1:50 internal floorplans and sections. The building was from 1600, so it was pretty wonky. Compared to using traditional methods - i.e. a disto and tape with a reflectorless total station, I think that the data would be perfectly fit for purpose.
On the basis it has already registered to 3mm, thereby confirming your target control accuracy, this would imply there is probably an incorrect constraint created by the "auto-align feature".
I agree, but it's possible that it was not an "auto-aligned" cloud but a manually aligned cloud. The weird thing is that I can't see the error when I slice the clouds. If I have the time later on, I will re-register the scans that used that target (there are only 2 of them) and see if the result can be improved.
"mean absolute error" does not negate local errors
Yeah, I should have mentioned the 11mm error at one of the targets in my summary. It's always tempting for the user to just disable a constraint that gives a poor result - I'd never do this without a valid reason (such as the target was jolted or the center cannot be established).

I always try to use 4-5 common targets per scan, which I think many would consider to be overkill. I would now consider using 2 targets plus C2C for projects where C2C is viable. Obviously scanning in a large open space is no good for C2C - also things like monitoring projects will need to be registered to permanent nails.
User avatar
Phill
V.I.P Member
V.I.P Member
Posts: 653
Joined: Wed Feb 06, 2008 7:19 am
16
Full Name: Phillip Nixon
Company Details: SKM
Company Position Title: Surveyor
Country: Australia
Location: Sydney
Has thanked: 7 times
Been thanked: 45 times

Re: Cloud Registration Case Study - Cyclone V9.0.2

Post by Phill »

James

Excellent write up again. I am wondering if you could expand, or maybe do a few checks for me as I have some questions.

1. When you say you used targets, were these aquired using external methods (ie. a total station) or did you just use "common targets" from scan to scan to perform your closed loop traverse?

2. If your were to disable your targets and just rely on the cloud to cloud (you might need to keep your targets for the first and last scan) how far out are the diableds targets say... at mid traverse.

3. How long in distance was your closed loop traverse?

The reason I ask is while it is something I have always wanted to check, I have never done a long line or long loop using cloud to cloud. We tend to rely on a total station to acquire the targets due to the inherient advantages of a total station (2 faces, direct pointing (rather than interpolated), longer range, better single distance accuracy, secondary check on scanner accuracy (we once had a scanner that had given itself a scale factor of 0.9996)). The problem that I saw with cloud to cloud was the propogation of error.

If you have a missmatch of say 3mm on your first cloud to cloud, on the second cloud to cloud this error would be 6mm, not taking into account any error in the second cloud to cloud, times this by 33 and it gets a little large. What you are in theory doing is taking a backsight the size of the distance between two targets in a single scan and measuring a foresight 33 times that distance. Of course you closed that loop, so that removes some of the error, but there is the possiblity the scan in the middle of the loop could be out by half of the closing error.

This is why I am wondering how the cloud to cloud compares to total station aquired targets, or even to common targets (which I think is what you have done).

Cheers

Phill
Mikolajczyk
I have made 50-60 posts
I have made 50-60 posts
Posts: 58
Joined: Tue Jan 24, 2012 2:49 pm
12
Full Name: Marcin Mikolajczyk
Company Details: MM Survey ApS
Company Position Title: Technical
Country: Denmark
Linkedin Profile: Yes

Re: Cloud Registration Case Study - Cyclone V9.0.2

Post by Mikolajczyk »

Phill,

I can answer your question but it was in cyclone 8. I made 20 scans in a loop and register first with second, second with third etc. Scan 1 and 20 were about meter appart.

In the registered pointcloud I turned on scan 1 and fit a pipe (approx 800mm diameter) then turned on scan 20 and did the same. The distance between centerlines of the same pipe on two scans was 9mm.

The overal project accuracy according to cyclone was 0.000 but as we can see, this can not be trusted.
I cant remember now but I think that each scan had 7million points and it was reather low quality.

After making auto-add cloud constrain registration was tighten and overal accuracy was 1 or 2 mm.

Marcin
Jamesrye
V.I.P Member
V.I.P Member
Posts: 643
Joined: Mon Aug 11, 2008 4:13 pm
15
Full Name: James Rye
Company Details: Merrett Survey Partnership
Company Position Title: Spatial Analyst
Has thanked: 28 times
Been thanked: 69 times

Re: Cloud Registration Case Study - Cyclone V9.0.2

Post by Jamesrye »

Phil,

"Common targets" from scan to scan were used to close the loop. Below shows the route - note that the building had some stairs that I scanned up to get to the second floor - then I closed out onto the same targets that I used at the start by opening a window.
route.jpg
Just to be clear, the work is a traverse in that it starts at a given point, runs in a circuit and eventually returns to the point of beginning. However, I did not use the traverse function on the C10 where you record backsight and foresight.
The problem that I saw with cloud to cloud was the propogation of error.
For sure this is a problem if you just use pairwise C2C, however - as my study showed, you can ask cyclone to "auto add cloud constraints" such that it looks at all scan overlaps. It improved my registration a lot - to be honest I'm not sure, but it might even count as "closing the loop". This should reduce/eliminate the propogation of error that you mentioned. If your project were "open" - not closed, then I think that you'd need targets for sure.

With regards to your question 2, the targets - brilliantly named as "aaa" and "inside" were roughly mid-way through the loop, so you can identify them in the registration screenshots in my previous post.

Some quick thoughts:
1. I could have spent longer adjusting the cloud constraints to better fit the targets - especially where we see the biggest error. But this would defeat the point of the exercise as the test was to see what accuracy I would achieve without using targets.
2. Someone else might achieve better results than me through more careful "initial alignment" of the scans in the 2D window.
3. Due to the dual axis compensator the vertical residual is "always" really good. I've used a scanner without said compensator and been amazed at how far out the errors have been prior to closing the loop.
4. It's hard to know how accurate the C2C is without using targets as a check - which sort of defeats the point. For sure you can get what looks like a great result only to find that the scans are misaligned. I see this as the biggest weakness and an area that could be improved upon.

What I am interested in is the implications that C2C has for workflow. Say you are scanning a large complex plant, my current workflow would be to use a total station to accurately coordinate a number of points. Then I'd plaster the place in Black and White targets and off I go. Is it now acceptable to just use 2 targets and C2C instead?

At the end of the day, I guess C2C is just another useful tool and it should be used where appropriate - the question is how we define appropriate. Certainly the C2C result that I achieved in my study would be appropriate for the type of survey that I was carrying out.
You do not have the required permissions to view the files attached to this post.
User avatar
Matt Young
Honorary Member
Honorary Member
Posts: 3929
Joined: Fri Nov 30, 2007 3:03 pm
16
Full Name: Matt Young
Company Details: Baker Hicks
Company Position Title: CAD-BIM Lead
Country: UK
Linkedin Profile: No
Has thanked: 45 times
Been thanked: 40 times

Re: Cloud Registration Case Study - Cyclone V9.0.2

Post by Matt Young »

Firstly, I have only had one attempt at the auto registration in cyclone 9.0 - the result was a screen and some message about no thumbnails created... I must go back to this at some point and see where I went wrong. I am glad to hear that someone is having a good experience using this new method. What I am most interested in is how the registration result is being reported? for the industrial work we do it is important that everything we do is reportable.

C2C seems to me to be quite robust if the amount of data for overlap is enough. One key thing to remember with C2C is that when you have run it you still need to go in to the constraints tab and either turn off some constraints or at least lower the weighting in order to let the other constraints work in your favour.

My currently preferred registration workflow is:

It's worth noting that I never really like the idea of using the traverse function on the scanner. It works well but in my opinion adds complication and complication adds room for error.

Full 360 scan every time (better to have it and not need it than need it and not have it)

At least three total station controlled targets for every scan (compensator always on) - you only need two targets but redundancy is useful. Co-incident vertices between scanner targets- up to you it works well with or without.

Register scans with total station control.

Edit any mistakes or mis-named targets using constellation comparison between control and scan control spaces (everyone makes mistakes from time to time)

Then turn off any constraints with errors higher than 3mm

Add scans that may be P2P (sometimes it's necessary)

Run C2C on all scans

Turn off or lower weights for constraints to release any remaining tension in the network.

Result - always 1 to 2mm or better

Physical check with limit box or slices in 3 axis - always seem very good.

Is auto registration better than that? maybe it's quicker? maybe it's easier? I won't know until I get time to really try it out properly.

I have more than a thousand project databases done in the above way and they are all registered to a degree that I can confidently say is more than fit for purpose.

The idea of a closed loop in my opinion does not apply to laser scanning in the same way as a total station traverse. The traditional traverse relies on angles and distances to single point locations. The laser scan typically records so much more information in 3D and the mesh fit between two scans is using far more information for the fit than one might think. Take a long line of single scans along a street with 3mm errors between each one, the 3mm error is spread across the whole street in three dimensions. I really do believe that if you compared a single point at each end of the street taken from a total station with the relative points in the scan, even with 3mm meshing errors it would be extremely close.
If you don't see that there is nothing, then you are kidding yourself.
Post Reply

Return to “Registration Software”