About ICP registration

Discuss all CloudCompare issues here.
Post Reply
mehmetvozdogan
I have made 10-20 posts
I have made 10-20 posts
Posts: 18
Joined: Fri Feb 01, 2019 9:49 pm
Full Name: Mehmet Volkan Ozdogan
Company Details: Dokuz Eylul University
Company Position Title: Research Assistant
Country: Turkey
Linkedin Profile: Yes

About ICP registration

Post by mehmetvozdogan » Tue Apr 09, 2019 7:01 pm

Hi,

I have some scan data and when i tried to register them with Fine registration (ICP) the RMS error is about 0.012 (final overlap %1 and random sampling imit is 300000) but i need more precise registration. I applied fine registration again with registered data and the RMS error decrease to 0.0072 is that right way for precising the registration.


Briefly;
DATA1 and DATA2 registered with ICP then i applied ICP again between DATA1 and DATA2.registered. And RMS error decreases significantly is this process correct for a high precise registration

Thank you..

User avatar
danielgm
V.I.P Member
V.I.P Member
Posts: 155
Joined: Thu Jun 20, 2013 10:45 am
Full Name: Daniel Girardeau-Montaut
Company Details: CloudCompare
Company Position Title: Administrator
Country: France
Linkedin Profile: Yes
Location: Grenoble, France
Contact:

Re: About ICP registration

Post by danielgm » Tue Apr 09, 2019 11:10 pm

1% percent overlap seems very low! Not sure the result is reliable...

Otherwise pay attention that the RMS is only computed on the points selected during the last iteration (points are randomly selected - up to 300000 here - but only inside the closest points defined thanks to the overlap parameter... so there might not be that much points).

Running the algorithm twice can help sometimes (in case ICP is 'stuck' in a local minima). But once again, you'd better have a reliable selection of points to avoid jumping to a wrong solution.
Daniel
CloudCompare admin
http://www.cloudcompare.org/

User avatar
stevenramsey
Global Moderator
Global Moderator
Posts: 1680
Joined: Sun Aug 12, 2007 9:22 pm
Full Name: Steven Ramsey
Company Details: 4DMax
Company Position Title: Technical Specialist Scanning
Country: UK
Skype Name: steven.ramsey
Linkedin Profile: Yes
Location: London
Contact:

Re: About ICP registration

Post by stevenramsey » Wed Apr 10, 2019 8:46 am

it will also depend on the scanner used. all scanners have inherent error each one is different as you can't say that a single scan has a zero error. even two scans from the same location will unlikely give you a RMS value much better than the scanners error. As Daniel says 1% is very very low for an overlap
Steven Ramsey

Home steven.ramsey@laserscanningforum.com
Work steven@4DMax.co.uk
Mobile +44 7766 310 915

mehmetvozdogan
I have made 10-20 posts
I have made 10-20 posts
Posts: 18
Joined: Fri Feb 01, 2019 9:49 pm
Full Name: Mehmet Volkan Ozdogan
Company Details: Dokuz Eylul University
Company Position Title: Research Assistant
Country: Turkey
Linkedin Profile: Yes

Re: About ICP registration

Post by mehmetvozdogan » Wed Apr 10, 2019 4:41 pm

danielgm wrote:
Tue Apr 09, 2019 11:10 pm
1% percent overlap seems very low! Not sure the result is reliable...

Otherwise pay attention that the RMS is only computed on the points selected during the last iteration (points are randomly selected - up to 300000 here - but only inside the closest points defined thanks to the overlap parameter... so there might not be that much points).

Running the algorithm twice can help sometimes (in case ICP is 'stuck' in a local minima). But once again, you'd better have a reliable selection of points to avoid jumping to a wrong solution.

Thank you for your answer,

What is the suggested overlap ratio for a reliabe registration (%10 ? )
And what can i do for a better registration?

Thank you

User avatar
danielgm
V.I.P Member
V.I.P Member
Posts: 155
Joined: Thu Jun 20, 2013 10:45 am
Full Name: Daniel Girardeau-Montaut
Company Details: CloudCompare
Company Position Title: Administrator
Country: France
Linkedin Profile: Yes
Location: Grenoble, France
Contact:

Re: About ICP registration

Post by danielgm » Wed Apr 10, 2019 8:20 pm

Yes, 10% is the minimum. And if you can increase that (during scanning) that's of course better. You'd better take overlapping parts of the scan that have some geometry and not too flat or smooth areas.
Daniel
CloudCompare admin
http://www.cloudcompare.org/

Post Reply

Return to “CloudCompare”