Cluster Registration and Lock vs Unlock

Discuss FARO hardware here.
Post Reply
BlueSasquatch
I have made <0 posts
I have made <0 posts
Posts: 4
Joined: Wed Jul 24, 2019 5:29 pm
4
Full Name: Chad Berens
Company Details: UCI
Company Position Title: Project Engineer
Country: USA
Linkedin Profile: No
Has thanked: 4 times

Cluster Registration and Lock vs Unlock

Post by BlueSasquatch »

Been scanning and processing for a few years, we use a Focus S120 and currently Scene 2019. Looking to improve my point-clouds, there is ghosting issues that are pretty consistent and on occasions maybe 1/40 scans is quite noticeably off.

I have a facility with 180 scans, broken down into 1st Floor (70), 2nd Floor (50), Roof (30)and Exterior (30)

About 1/2 of the scans are 1/2 Resolution x2 Quality w/photos
The other half are 1/4 Resolution x3 Quality w/no photos.

I pulled in all of the files, pre-processed them and pre-clustered them based on the four main divisions, listed above.

Processing
Typical Processing Settings:
• Create Scan Point Clouds: Yes
• Skip Fully Processed Scans: Yes
• Colorization: Yes
• Filters: Yes (Threshold 200)
• Distance Filter: Yes (0-656ft)
• Stray Point Filter: Yes (3px , 0.07ft, 50%)
• Edge Artifact Filter: Yes
• Find Targets: Find Spheres (If Applicable)
• Perform Automatic Registration: No

I looked up the S120, that has a range of 0.6m-120m. Which is roughly 400 feet, so it is clear that my distance filter could be improved by making this at a minimum, 400 feet rather than 656 feet (default)

Is there a better distance to set this filter at? I understand accuracy of the scanner drops, the greater the distance, and it will depend on the tolerances we are comfortable with. However the building we scanned has 225 feet as the longest side, which means at 400 feet, the vast majority of our scans have the potential to have shared points from all of the scans (if there were no obstructions, theoretically)

Also what is the difference between Processing Distance Filter and Point Cloud Distance Filter? Is the point-cloud distance filter just there, if you didn't use the distance filter in processing? Or wish to narrow your distance later? If you were to make both distance filter numbers the same, there would be no change, correct?

Registration
In a broad sense, My process is to get all the scans laid out correctly with a quick-correspondence view, and then "refine" the registration by using Cloud to Cloud, which I typically do multiple times; Default of 32ft, then I re-attempt at 3 feet and then at 0.3 feet, In other jobs I've gone as far as .01 feet. Ideally each time, I look at the Overall Statistics % under the Scan manager - Scan Point Tension tab, and shoot for an improvement (a higher number) each time. Occasionally the number gets worse and I revert back to the previous save, and consider that cluster finished. (I know that isn't the best metric to use, open to suggestions, we are self taught here)

This project used targets, so I attempt an automatic reg based on targets, I get clusters within my clusters now, and by either manual registration (I actually quite like this option) or Visual via correspondence view, I have gotten the four clusters to appear correct, or within initial placement/reason. I've got green lights on my overview for registration, the scans look pretty good as I shuffle through them.

What level of inspection and how do you all go about it, on the registration step? Do you look very close at each and every scan? I can tell in correspondence top view, when a scan is roughly in the right spot, and I thought the cloud to cloud would tighten them up, and it usually does, but when I end up in Recap/Autocad and we begin to model from the point cloud, we notice sometimes that a scan is very poorly registered. I've found it very hard to notice these issues, when in scene. Do I just need to take more time to inspect things closer?
Can I use the Registration Report to note scans that are red, and visually check them one at a time? I've been hunting all week for some video or post to explain exactly what I'm looking at with the registration report, and what to do to improve it. Some forum posts say that it's garbage, but my boss doesn't like that excuse, I know that red must be bad and green is good, but hell if I follow much more of it outside that.

Image

Image


As for the subject of the thead: When I have 1st and 2nd floor, clustered and registered correctly within themselves, and I want to now "combine" the two floors, I've got some decent overlap with doorways, stairways, chunks of the floor gone, etc. Should I lock the clusters before registering them together? My team seems to be in the camp of getting them close, locked, but when refining them, to unlock it so the scans can all move a little if needed, to get the best fit. The worry we have is if they are both locked, that you can get a 2nd and 1st floor that are correct on one end, but wrong on the farther end, Im not explaining this all too well, but can't seem to find a good argument for one way or the other.


Point Cloud Creation
These are the settings we typically stick with.
  •  Eliminate Duplicate Points: Yes
     Search Radius: Medium
     Close Surfaces: Yes
     Full Color Detail: No
     Homogenize Point Density: No
     Apply Color Balancing: No
     Distance Filter: 90m
Again, I question the distance filter here, see previous paragraph on it.

Also would like to complain about the "disk space needed" portion of the point cloud creation window, mine will frequently tell me it's got enough space yet fails after 15 hours of processing. We are grabbing a new 2TB SSD to remedy this, but it's bitten me in the ass a few times, I've found FARO websites suggesting a mix of 2-4 and like a 4-7x free space of your scan data files, so we are flying with a 11x scan data space needed, but that just seems like such a vague answer. Also which file size is this based off of? My FLS Scans on the scanner SD card is smaller than the FLS scans in the scene folders, and is that even the correct spot to pull the size from?


I can post some pictures of the project in particular in a bit, let me grab some useful screen shots.
ericguizzetti
V.I.P Member
V.I.P Member
Posts: 205
Joined: Tue Mar 06, 2018 6:56 pm
6
Full Name: Eric Guizzetti
Company Details: Construction and Engineering
Company Position Title: RealityCapture
Country: USA
Linkedin Profile: Yes
Has thanked: 14 times
Been thanked: 11 times
Contact:

Re: Cluster Registration and Lock vs Unlock

Post by ericguizzetti »

Get an RTC360, problem solved.
BlueSasquatch wrote: Fri Oct 15, 2021 7:32 pm Been scanning and processing for a few years, we use a Focus S120 and currently Scene 2019. Looking to improve my point-clouds, there is ghosting issues that are pretty consistent and on occasions maybe 1/40 scans is quite noticeably off.

I have a facility with 180 scans, broken down into 1st Floor (70), 2nd Floor (50), Roof (30)and Exterior (30)

About 1/2 of the scans are 1/2 Resolution x2 Quality w/photos
The other half are 1/4 Resolution x3 Quality w/no photos.

I pulled in all of the files, pre-processed them and pre-clustered them based on the four main divisions, listed above.

Processing
Typical Processing Settings:
• Create Scan Point Clouds: Yes
• Skip Fully Processed Scans: Yes
• Colorization: Yes
• Filters: Yes (Threshold 200)
• Distance Filter: Yes (0-656ft)
• Stray Point Filter: Yes (3px , 0.07ft, 50%)
• Edge Artifact Filter: Yes
• Find Targets: Find Spheres (If Applicable)
• Perform Automatic Registration: No

I looked up the S120, that has a range of 0.6m-120m. Which is roughly 400 feet, so it is clear that my distance filter could be improved by making this at a minimum, 400 feet rather than 656 feet (default)

Is there a better distance to set this filter at? I understand accuracy of the scanner drops, the greater the distance, and it will depend on the tolerances we are comfortable with. However the building we scanned has 225 feet as the longest side, which means at 400 feet, the vast majority of our scans have the potential to have shared points from all of the scans (if there were no obstructions, theoretically)

Also what is the difference between Processing Distance Filter and Point Cloud Distance Filter? Is the point-cloud distance filter just there, if you didn't use the distance filter in processing? Or wish to narrow your distance later? If you were to make both distance filter numbers the same, there would be no change, correct?

Registration
In a broad sense, My process is to get all the scans laid out correctly with a quick-correspondence view, and then "refine" the registration by using Cloud to Cloud, which I typically do multiple times; Default of 32ft, then I re-attempt at 3 feet and then at 0.3 feet, In other jobs I've gone as far as .01 feet. Ideally each time, I look at the Overall Statistics % under the Scan manager - Scan Point Tension tab, and shoot for an improvement (a higher number) each time. Occasionally the number gets worse and I revert back to the previous save, and consider that cluster finished. (I know that isn't the best metric to use, open to suggestions, we are self taught here)

This project used targets, so I attempt an automatic reg based on targets, I get clusters within my clusters now, and by either manual registration (I actually quite like this option) or Visual via correspondence view, I have gotten the four clusters to appear correct, or within initial placement/reason. I've got green lights on my overview for registration, the scans look pretty good as I shuffle through them.

What level of inspection and how do you all go about it, on the registration step? Do you look very close at each and every scan? I can tell in correspondence top view, when a scan is roughly in the right spot, and I thought the cloud to cloud would tighten them up, and it usually does, but when I end up in Recap/Autocad and we begin to model from the point cloud, we notice sometimes that a scan is very poorly registered. I've found it very hard to notice these issues, when in scene. Do I just need to take more time to inspect things closer?
Can I use the Registration Report to note scans that are red, and visually check them one at a time? I've been hunting all week for some video or post to explain exactly what I'm looking at with the registration report, and what to do to improve it. Some forum posts say that it's garbage, but my boss doesn't like that excuse, I know that red must be bad and green is good, but hell if I follow much more of it outside that.

Image

Image


As for the subject of the thead: When I have 1st and 2nd floor, clustered and registered correctly within themselves, and I want to now "combine" the two floors, I've got some decent overlap with doorways, stairways, chunks of the floor gone, etc. Should I lock the clusters before registering them together? My team seems to be in the camp of getting them close, locked, but when refining them, to unlock it so the scans can all move a little if needed, to get the best fit. The worry we have is if they are both locked, that you can get a 2nd and 1st floor that are correct on one end, but wrong on the farther end, Im not explaining this all too well, but can't seem to find a good argument for one way or the other.


Point Cloud Creation
These are the settings we typically stick with.
  •  Eliminate Duplicate Points: Yes
     Search Radius: Medium
     Close Surfaces: Yes
     Full Color Detail: No
     Homogenize Point Density: No
     Apply Color Balancing: No
     Distance Filter: 90m
Again, I question the distance filter here, see previous paragraph on it.

Also would like to complain about the "disk space needed" portion of the point cloud creation window, mine will frequently tell me it's got enough space yet fails after 15 hours of processing. We are grabbing a new 2TB SSD to remedy this, but it's bitten me in the ass a few times, I've found FARO websites suggesting a mix of 2-4 and like a 4-7x free space of your scan data files, so we are flying with a 11x scan data space needed, but that just seems like such a vague answer. Also which file size is this based off of? My FLS Scans on the scanner SD card is smaller than the FLS scans in the scene folders, and is that even the correct spot to pull the size from?


I can post some pictures of the project in particular in a bit, let me grab some useful screen shots.
User avatar
Leandre Robitaille
V.I.P Member
V.I.P Member
Posts: 542
Joined: Sat Aug 03, 2019 1:53 am
4
Full Name: Leandre Robitaille
Company Details: Cima+
Company Position Title: Civil Technician - Surveyor
Country: Canada
Has thanked: 61 times
Been thanked: 246 times

Re: Cluster Registration and Lock vs Unlock

Post by Leandre Robitaille »

ericguizzetti, that has to be the most unhelpfull comment I have seen in a long time...


BlueSasquatch. this is a deeper issue within your scene project. This error occurs when doing target based registrations and not ''closing'' the loops. Make sure you close loops, you could do a c2c of 2 scans that have no spheres between them to ''close a loop''
View the scans in correspondance view to identify where this error commes from
Heck, I would even be ready to offer you a free call threw a team meeting (I am only available this saturday 23 oct, buisiess time of the year for me...and we can look over your project together)

However when figuring out the faulting scan and why your loop is not closing you should be able to fix this, don't be afraid to use clusters to aid your loop. You can c2c 2 scans together and ''lock'' that cluster. Make sure to bring all scans on the same clusterlevel. Dont lock a cluster per level or your error wont spread...this is when you see these offsets.
User avatar
Tanguy Nédélec
I have made 90-100 posts
I have made 90-100 posts
Posts: 99
Joined: Fri Oct 23, 2015 3:15 pm
8
Full Name: NEDELEC Tanguy
Company Details: Prigent et Associes
Company Position Title: Surveyor - nooks and crannies explorer
Country: France
Linkedin Profile: No
Location: Dinan
Has thanked: 7 times
Been thanked: 19 times

Re: Cluster Registration and Lock vs Unlock

Post by Tanguy Nédélec »

Hello and welcome to the forum (well, for your first post anyway).

I have a lot of experience with Scene. The only way to ensure a good accuracy with C2C is to check every cluster with thin clipping boxes (do one horizontal and a few vertical ones).

When there are mismatches like you mentioned, you have to subcluster, i.e. divide a bigger cluster into smaller fragments. Registration or preprocessing parameters don't seem to matter much. Sometimes trying a second C2C attempt after the first one gets things into place, but not always.

Don't trust the statistics either, they often don't reflect the reality.

About locking clusters : can't really say what are the consequences of registering two unlocked clusters as I never do this. If you follow the clipping box / subcluster method your individual clusters should be well registered, so locking them shouldn't show major inaccuracies in the end.

On a side note: your project seems to be in a difficult environment for C2C, maybe putting a few targets and measuring them with a total station would help? Like decide in advance of a few reasonably sized clusters, and put 4-5 well distributed targets so that each cluster can be registered independently on these references. That's usually a good methodology for bigger projects, and it has the advantage that it gives you immediatly measurable errors.

I've only answered your questions I'm most familiar with, I will let someone else answer the others. Hope this helps.
BlueSasquatch
I have made <0 posts
I have made <0 posts
Posts: 4
Joined: Wed Jul 24, 2019 5:29 pm
4
Full Name: Chad Berens
Company Details: UCI
Company Position Title: Project Engineer
Country: USA
Linkedin Profile: No
Has thanked: 4 times

Re: Cluster Registration and Lock vs Unlock

Post by BlueSasquatch »

tengui wrote: Mon Oct 18, 2021 10:36 am On a side note: your project seems to be in a difficult environment for C2C.
Thank you for your post, our office is in a debate on the usefulness of locking clusters before aligning these clusters, and leaving them all unlocked. Everyone else seems to be in the camp of unlocking them to allow them to make minor changes as needed? For instance the 1st and 2nd floor of this facility has a concrete floor, so there is not that much overlap, yet there are areas where there is no floor, or a stairwell, etc, and we may get overlap randomly throughout the floor. If we get say the 2nd floor registered well, and then go to align/register it with the 1st floor, there may be a handful of potential overlap/connecting data points, on the far ends of the floors. If the clusters are unlocked, our reasoning is it would allow for a better connection to the 1st floor, where there is overlap, rather than being locked and possibly causing good connection to the 1st floor in one spot, and poor connection in the other?

Most of our work is done at processing facilities with miles of pipe, steam, tanks, equipment, people, etc. We almost always use spheres, to get an initial registration and then follow up with C2C for refinement, starting to read that maybe we skip the C2C if our target distribution was good?
Leandre Robitaille wrote: Mon Oct 18, 2021 4:40 am BlueSasquatch. this is a deeper issue within your scene project. This error occurs when doing target based registrations and not ''closing'' the loops. Make sure you close loops, you could do a c2c of 2 scans that have no spheres between them to ''close a loop''
View the scans in correspondance view to identify where this error commes from

However when figuring out the faulting scan and why your loop is not closing you should be able to fix this, don't be afraid to use clusters to aid your loop. You can c2c 2 scans together and ''lock'' that cluster. Make sure to bring all scans on the same clusterlevel. Dont lock a cluster per level or your error wont spread...this is when you see these offsets.
Thank you for the offer and the help! This project was quite large and made a "loop" tricky, I do try to close a loop if I know I am going to be doing a loop, but don't always have a spare sphere to leave behind. For this project I moved in a S pattern over and over, which did allow me to utilize many of the same spheres, I feel like the sphere distribution was overall pretty good.

You are suggesting that I lock clusters as they look correct, once inspected?

What are good/bad scan tensions to shoot for? I know the numbers aren't as good as the eye but just as a starting point on what to look at? I can have a green registration but a scan tension of .302 for instance.
badam
V.I.P Member
V.I.P Member
Posts: 916
Joined: Tue May 11, 2021 5:36 pm
2
Full Name: Adam Berta
Company Details: InnoScan 3D Hungary Kft
Company Position Title: unknown
Country: Hungary
Linkedin Profile: No
Has thanked: 52 times
Been thanked: 297 times
Contact:

Re: Cluster Registration and Lock vs Unlock

Post by badam »

BlueSasquatch wrote: Mon Oct 18, 2021 3:01 pm
Thank you for the offer and the help! This project was quite large and made a "loop" tricky, I do try to close a loop if I know I am going to be doing a loop, but don't always have a spare sphere to leave behind. For this project I moved in a S pattern over and over, which did allow me to utilize many of the same spheres, I feel like the sphere distribution was overall pretty good.

You are suggesting that I lock clusters as they look correct, once inspected?

What are good/bad scan tensions to shoot for? I know the numbers aren't as good as the eye but just as a starting point on what to look at? I can have a green registration but a scan tension of .302 for instance.
Loop closure is key especialy in these kind of situtions. Maybe worth to try to do a loop closure with cloud to cloud, or at least some kind of manual targets. That is why i like cyclone (register 360) products you have plenty of tools to create c2c constraints, and you will not have to use targets for good registration. I wouldn't lock the "clusters". after loop closure you may get some feedback if one of the constrains was wrong beforehand.
BlueSasquatch
I have made <0 posts
I have made <0 posts
Posts: 4
Joined: Wed Jul 24, 2019 5:29 pm
4
Full Name: Chad Berens
Company Details: UCI
Company Position Title: Project Engineer
Country: USA
Linkedin Profile: No
Has thanked: 4 times

Re: Cluster Registration and Lock vs Unlock

Post by BlueSasquatch »

badam wrote: Mon Oct 18, 2021 3:46 pm Loop closure is key especialy in these kind of situtions. Maybe worth to try to do a loop closure with cloud to cloud, or at least some kind of manual targets. That is why i like cyclone (register 360) products you have plenty of tools to create c2c constraints, and you will not have to use targets for good registration. I wouldn't lock the "clusters". after loop closure you may get some feedback if one of the constrains was wrong beforehand.
I understand the loop closure concept as covered in some Faro manuals / videos and how a circle can appear like a spiral, if one does not loop close. I'm having a harder time with the concept of a loop closure or how to go about it, when looking at multiple floors. Take the 2nd floor and the roof for instance, I only have 1 doorway to connect the two areas. Theoretically the 2nd is connected to the first and then exterior so the roof has some overlap with the ground scans, so there is an over-all arching loop.

How do you loop closure with cloud to cloud? I will often manually mark planes between scans if I can't get target to target to work, with the spheres, and I watched a video on loop closure with this method included.

I've split the roof up into 3 sub-clusters; Upper, Lower and the Tank Room, where there are five covered tanks, the access is merely from the roof, but it's enclosed as a room itself. I have successful target to target registration. I then move to C2C at 1 ft search distances and I'm getting a Scan Point Tension Overall % of 65-75 which seems quite good, all the MEAN (in) numbers are below 0.2 and from what I can see, zooming in on the slices/topviews, everything looks lined up to me. Is there another step I should be doing? When I run C2C at 0.1 ft, the numbers get worse, so I revert back to the previous save.

When you have the scan-manager pop up, after an operation, there are four buttons at the bottom; Get, Apply, Okay and Cancel. What does Get or Apply do? I've always just hit Okay. Today I was curious and hit Apply, it churned for a while and then popped back up, with worse numbers. Is this akin to hitting enter over and over again on your calculator, where it applies the same settings, again? Why would the numbers change?
User avatar
landmeterbeuckx
V.I.P Member
V.I.P Member
Posts: 1616
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: Cluster Registration and Lock vs Unlock

Post by landmeterbeuckx »

BlueSasquatch wrote: Mon Oct 18, 2021 6:51 pm
badam wrote: Mon Oct 18, 2021 3:46 pm Loop closure is key especialy in these kind of situtions. Maybe worth to try to do a loop closure with cloud to cloud, or at least some kind of manual targets. That is why i like cyclone (register 360) products you have plenty of tools to create c2c constraints, and you will not have to use targets for good registration. I wouldn't lock the "clusters". after loop closure you may get some feedback if one of the constrains was wrong beforehand.
I understand the loop closure concept as covered in some Faro manuals / videos and how a circle can appear like a spiral, if one does not loop close. I'm having a harder time with the concept of a loop closure or how to go about it, when looking at multiple floors. Take the 2nd floor and the roof for instance, I only have 1 doorway to connect the two areas. Theoretically the 2nd is connected to the first and then exterior so the roof has some overlap with the ground scans, so there is an over-all arching loop.

How do you loop closure with cloud to cloud? I will often manually mark planes between scans if I can't get target to target to work, with the spheres, and I watched a video on loop closure with this method included.

I've split the roof up into 3 sub-clusters; Upper, Lower and the Tank Room, where there are five covered tanks, the access is merely from the roof, but it's enclosed as a room itself. I have successful target to target registration. I then move to C2C at 1 ft search distances and I'm getting a Scan Point Tension Overall % of 65-75 which seems quite good, all the MEAN (in) numbers are below 0.2 and from what I can see, zooming in on the slices/topviews, everything looks lined up to me. Is there another step I should be doing? When I run C2C at 0.1 ft, the numbers get worse, so I revert back to the previous save.

When you have the scan-manager pop up, after an operation, there are four buttons at the bottom; Get, Apply, Okay and Cancel. What does Get or Apply do? I've always just hit Okay. Today I was curious and hit Apply, it churned for a while and then popped back up, with worse numbers. Is this akin to hitting enter over and over again on your calculator, where it applies the same settings, again? Why would the numbers change?
Colorize after registration, not before. This is already often told.
LSBbvba
Surveying services - 3D Laserscanning
Tel : +32477753126
www.lsbbvba.be
[email protected]
BlueSasquatch
I have made <0 posts
I have made <0 posts
Posts: 4
Joined: Wed Jul 24, 2019 5:29 pm
4
Full Name: Chad Berens
Company Details: UCI
Company Position Title: Project Engineer
Country: USA
Linkedin Profile: No
Has thanked: 4 times

Re: Cluster Registration and Lock vs Unlock

Post by BlueSasquatch »

landmeterbeuckx wrote: Mon Oct 18, 2021 7:00 pm

Colorize after registration, not before. This is already often told.
How do you apply color after registration? Im using 2019 scene with that new ribbon interface. Also why do it afterwards?

I tried to take this project and re-process the scans, to save re-registering them, but to adjust some of the filter settings, kept getting errors and corrupt scans after the process itself took 12+ hours. Is there a way to do filters again, afterwards? Or just one of the filters?
User avatar
TommyMaddox
V.I.P Member
V.I.P Member
Posts: 514
Joined: Mon Apr 25, 2016 7:15 pm
7
Full Name: Tommy R Maddox III
Company Details: ONSITE3D
Company Position Title: Director of Technology
Country: USA
Linkedin Profile: Yes
Location: Calgary, Alberta
Has thanked: 67 times
Been thanked: 130 times

Re: Cluster Registration and Lock vs Unlock

Post by TommyMaddox »

When you have limited overlap connection zones you first need to make clusters that contain your stairwell(s)/outside shots to/through windows and doors, etc... to form a skeleton to which the larger areas are attached.

Locking clusters is your friend as soon as you have verified the registration quality of sections. Clipping boxes and detailed visual QAQC in correspondence view is mandatory especially when you don't have extensive targeting and totalstation control to check against.

Lastly, setup some checkerboard targets in your office to scan with your old S120. Because it can't do the OnSite Compensation routine that the newer scanners have, and you don't have extensive total station control on projects, you'll need to have some way to validate the accuracy of the unit. Measuring the distance between sets of the same checkerboards scanned from different positions, and closely examining the individual scans for seam-line mismatch will help you validate that the unit is performing acceptably to an extent (obviously would be better to scan spheres that have been shot in with a total station or laser tracker).

If you need some training, I am available for consulting. The folks on the forum here can also be quite helpful.

EDIT: I will also add that you should never create a project pointcloud prior to completely finalizing both your registration and your alignment. I would also not do this before any manual cleaning is completed. Any updates you make to those categories will not carry through to the project pointcloud, and it must be deleted and recreated after those changes are made.
Post Reply

Return to “FARO”