Faro Scene - poor accuracy for sphere locations

To chat about anything else.
Post Reply
fobos8
Forum Supporter
Forum Supporter
Posts: 145
Joined: Sat Jul 07, 2018 9:19 pm
Full Name: Andrew
Company Details: Medina Surveys
Company Position Title: Surveyor
Country: Uk
Linkedin Profile: No
Has thanked: 16 times
Been thanked: 12 times

Faro Scene - poor accuracy for sphere locations

Post by fobos8 »

Hi guys

I've noticed that in my projects there is poor accuracy on the location of spheres. I ran a test. I scanned a geometrically good scene 10 metres to the left of a sphere and then scanned 10 metres to the right of the same sphere. I registered the two scans together using c2c and of course got great registration stats. However, the co-ordinates for the sphere differ by 7mm!

I've checked I'm using the correct sphere diameter. I ran the test on 1/4 res and x 2 quality. My scanner is about 4 months old and is a Focus 150 plus.
I'm getting Scene to auto select spheres during processing.

Please can someone advise me?

Kind regards, Andrew
Leandre Robitaille
I have made 90-100 posts
I have made 90-100 posts
Posts: 92
Joined: Sat Aug 03, 2019 1:53 am
Full Name: Leandre Robitaille
Company Details: Cima+
Company Position Title: Civil Technician
Country: Canada
Has thanked: 5 times
Been thanked: 13 times

Re: Faro Scene - poor accuracy for sphere locations

Post by Leandre Robitaille »

What sphere diameter are you using?
Who made those spheres?
How many points are on those spheres ?(aim for 1000+)
Are they in grass, if you manually detect the sphere does the yellow splash go on the floor or stay on the sphere?


Make sure the scanner seam does not start and close on a sphere.
Spheres cant be split (for exemple threw a fence, etc), even if you see 40% of the sphere on both side of a split,for exemple a branch ,only one half will be used in the detection.

We are missing information on your setup, if you are getting 7mm tension between 2 scans ,then your scanner might be off calibration...or you might just be doing something wrong here
fobos8
Forum Supporter
Forum Supporter
Posts: 145
Joined: Sat Jul 07, 2018 9:19 pm
Full Name: Andrew
Company Details: Medina Surveys
Company Position Title: Surveyor
Country: Uk
Linkedin Profile: No
Has thanked: 16 times
Been thanked: 12 times

Re: Faro Scene - poor accuracy for sphere locations

Post by fobos8 »

Leandre Robitaille wrote: Thu Aug 06, 2020 11:53 pm What sphere diameter are you using?
Who made those spheres?
How many points are on those spheres ?(aim for 1000+)
Are they in grass, if you manually detect the sphere does the yellow splash go on the floor or stay on the sphere?
Leandre, Thanks for your reply.

I just checked a different set of scans with a tighter registration so I can discount the registration error which is 1.5mm on the 3 scans I just checked. They are all within 12 metres of each other and the co-ordinates for the same sphere differ by up to 6mm. In two of the scans the number of points on the sphere is 1000+ and the third its 300 points. I thought the magic number was 80 points?

For the two scans that have 1000+ points on the sphere the co-ordinates of the sphere are showing to differ by 6mm. The distance to the sphere from these scans is 8metres and the 2 scan positions are 12 metres apart.

The spheres are 145mm dia, bought from Optical in the UK. They look like the same ones sold at Laserscanning Europe. The spheres are ontop of a tripod.

Kind regards, Andrew
VXGrid
V.I.P Member
V.I.P Member
Posts: 179
Joined: Fri Feb 24, 2017 10:47 am
Full Name: Martin Graner
Company Details: PointCab GmbH
Company Position Title: Research and Development
Country: Germany
Linkedin Profile: No
Has thanked: 22 times
Been thanked: 27 times
Contact:

Re: Faro Scene - poor accuracy for sphere locations

Post by VXGrid »

Hi Andrew,

just curious, but might it be possible that the issue is in the c2c rather than the spheres?
If it is feasable you can upload the data and somebody can have a look?


Martin
fobos8
Forum Supporter
Forum Supporter
Posts: 145
Joined: Sat Jul 07, 2018 9:19 pm
Full Name: Andrew
Company Details: Medina Surveys
Company Position Title: Surveyor
Country: Uk
Linkedin Profile: No
Has thanked: 16 times
Been thanked: 12 times

Re: Faro Scene - poor accuracy for sphere locations

Post by fobos8 »

Martin et al

Here's a link to the scans. If anyone wants to have a look that would be amazing.

https://1drv.ms/u/s!AicVK0yuchiNgx0yOQB ... F?e=rcDm2Z
max72
V.I.P Member
V.I.P Member
Posts: 739
Joined: Tue Feb 26, 2013 9:32 pm
Full Name: Massimo De Marchi
Company Details: Massimo De Marchi
Company Position Title: freelancer
Country: Italy
Skype Name: massimo_de_marchi
Has thanked: 1 time
Been thanked: 3 times
Contact:

Re: Faro Scene - poor accuracy for sphere locations

Post by max72 »

Probably I have a problem with the SDK. They imported as unregistered..
I'll try with scene LT.
Massimo

EDIT: also with scene LT no luck. They are roughly placed but not registered (1-2 meters off).
ing. Massimo De Marchi - +39 347 32 17 049 - www.studiodemarchi.net
fobos8
Forum Supporter
Forum Supporter
Posts: 145
Joined: Sat Jul 07, 2018 9:19 pm
Full Name: Andrew
Company Details: Medina Surveys
Company Position Title: Surveyor
Country: Uk
Linkedin Profile: No
Has thanked: 16 times
Been thanked: 12 times

Re: Faro Scene - poor accuracy for sphere locations

Post by fobos8 »

max72 wrote: Fri Aug 07, 2020 1:54 pm Probably I have a problem with the SDK. They imported as unregistered..
I'll try with scene LT.
Massimo

EDIT: also with scene LT no luck. They are roughly placed but not registered (1-2 meters off).
sorry guys - I uploaded the unregistered scans

Try this link. There are registered fls, registered e57 and raw files


https://1drv.ms/u/s!AicVK0yuchiNgx0yOQB ... F?e=n8zEkM
christopherbyrne18
V.I.P Member
V.I.P Member
Posts: 360
Joined: Thu Jan 17, 2008 8:18 pm
Full Name: Christopher Byrne
Company Details: Murphy Surveys
Company Position Title: Geospatial Manager
Country: Ireland
Linkedin Profile: Yes
Has thanked: 3 times
Been thanked: 1 time

Re: Faro Scene - poor accuracy for sphere locations

Post by christopherbyrne18 »

Hi,

I had a quick look in Cyclone.... Registerd C2C using the alignment tool and old school pick points. To me the data from the scanner is poor. Was it levelled correctly for each scan and not knocked or interfered with while scanning? Its also very noisy. Has it been calibrated recently?
Sphers.jpg
Sphere Info.jpg
Registration.jpg
Noisey Data 1.jpg
Noisey Data 2.jpg
Calibration.jpg
Sphere Target Co-ords.jpg
You do not have the required permissions to view the files attached to this post.
User avatar
Daniel Wujanz
I have made 100> posts
I have made 100> posts
Posts: 108
Joined: Fri Aug 24, 2018 11:26 am
Full Name: Daniel Wujanz
Company Details: technet GmbH
Company Position Title: 3D Laser Scanning Specialist
Country: Germany
Linkedin Profile: Yes
Location: Berlin
Has thanked: 44 times
Been thanked: 41 times
Contact:

Re: Faro Scene - poor accuracy for sphere locations

Post by Daniel Wujanz »

Dear Andrew,

me and Martin from PointCab just had a cheeky look at your data.

Our processing steps:

- Sphere detection in PointCab
- Plane detection in the scans (all following steps were done in Scantra 2.4)
- Plane based pairwise registration (in order to rule out that your C2C registration was erroneous)
- Block adjustment with three registrations and three inclinometer readings (assumed accuracy 0.3 mrad)

Outcome:
- Residuals of registration parameters after the block adjustment:
-station 5 to 7: 0.6 mm (3D-misclosure)
-station 5 to 6: 0.3 mm (3D-misclosure)
-station 6 to 7: 1.1 mm (3D-misclosure)

Note that this approach is indepedent to the applied registration procedure. I've explained the principle in the PointCab-webinar series:

https://www.youtube.com/watch?v=u_CWNKjD_vM&t=1505s

The general principle can be found around 16:00 minutes runtime.

Then we had a look at Leandre's suggestion that the scanner's calibration may be off. Scantra computes a priori values for the assumed precision of a registration and compares these values to the final outcome after the block adjustment (also, have a look at the webinar).

A simple example: you've measured a distance ten times with a Leica Disto and received a standard deviation of 5 mm for the mean range. Then you have to compare this value to the specified accuracy of the applied device. Assuming that this would be 1 mm, you'll end up with a signal to noise ratio (or empirical standard deviation as we surveyors / geodesists call it) of 5 mm / 1 mm = 5. This means your Disto has performed five times worse than expected and is hence a good reason for a stiff letter of complaint.

Back to your scanner: We've received an empirical standard deviation of 1.31 which tells us that everything looks fine at first glance. However, the sample is of course very small since we only have three scans. Indoor scans would have been preferable, but that's another story. 0.85 confirms the assumed accuracy of the inclinometer readings, so everything also appears to be fine.

Statistics
-------------------------------------------------------------------------------------
Observation Group sigma_0 Redundancy vtPv
-------------------------------------------------------------------------------------
total 1.16 10.0 100 13.37

transformations 1.31 6.1 61 10.56
vertical axes 0.85 3.9 39 2.81

Now to your spheres: We took the registration parameters from Scantra, imported them into PointCab and applied them to the spheres. Here's the outcome

Front005 -3.9171 -9.3034 -0.7551
Front006 -3.9182 -9.3027 -0.7534
Front007 -3.9092 -9.3028 -0.7537

The first two target locations coincide quite well while the third one looks funny - as you've initally mentioned. Since we know that our registrations are fine we assume that the spheres are to blame.

Hence, we intersected the sphere's mount on your tripod which looked suspicious. Then we had a look at another part of the point cloud (cross section through a car with 2 mm thickness) where everything fitted like a glove.

Sherlock Graner and Daniel Watson thus assume that the tripod has moved during the three scans. What makes us believe that? We finally had a look at your setup and noted that the tripod was partly placed on the lawn (1 leg) AND a sealed surface (the other 2 legs) - and that's something you should avoid at all cost.

All the best and have a good weekend

Martin and Daniel

P.S. We've attached the computed transformation parameters, so that you can apply them to your data. These are homogeneous coordinates (4 x 4 matrix) that e.g. CloudCompare, PointCab, Z+F LaserControl and most others should be able to interpret.

P.P.S. Thank you very much for sharing your data! A lot of people ask for help but do not upload any scans so it's guessing most of the time...
sim.herrod
V.I.P Member
V.I.P Member
Posts: 173
Joined: Thu Jan 14, 2016 12:26 pm
Full Name: Simeon Herrod
Company Details: Terra Measurement Limited
Company Position Title: 3D Data Manager
Country: UK
Linkedin Profile: No
Has thanked: 17 times
Been thanked: 27 times

Re: Faro Scene - poor accuracy for sphere locations

Post by sim.herrod »

Andrew, run the On-Site Compensation on your scanner.

https://knowledge.faro.com/Hardware/3D_ ... s_S_Series

Also, do you level every scan, and then make sure your tripod legs are tight?
Post Reply

Return to “General Chat”