Page 1 of 1
Fine registration ICP Normals
Posted: Sat Mar 05, 2022 4:13 am
by Lentep
How does selecting the Normals option improve the ICP results? If I choose "Same side", what is it changing in the calculations?
Re: Fine registration ICP Normals
Posted: Sat Mar 05, 2022 11:14 am
by daniel
If both your clouds have normals, then the points that have similar normals (and in this case not only the direction but also the orientation) will have more weight in the global registration process.
Re: Fine registration ICP Normals
Posted: Sat Mar 05, 2022 11:56 pm
by Lentep
daniel wrote: ↑Sat Mar 05, 2022 11:14 am
If both your clouds have normals, then the points that have similar normals (and in this case not only the direction but also the orientation) will have more weight in the global registration process.
If model and data have two points:
Data A - normal (1, 1, 1)
Data B - normal (0, 0.5, 0.5)
Model A - normal (1, 1, 1)
Model B - normal (1, 1, 1)
A is closest to A, B is closest to B
This means the process try to minimize distance between points A? And distance between points B is lower priority
Re: Fine registration ICP Normals
Posted: Mon Mar 07, 2022 1:45 pm
by daniel
Yes, exactly!
Re: Fine registration ICP Normals
Posted: Mon Mar 07, 2022 1:52 pm
by daniel
But actually, since ICP can potentially change the pairing at each iteration (since it pairs the nearest points, but this changes over time), this means that CC will consider that if Model B (1, 1, 1) and Data C (0.5, 0.5, 0.5) are matched, then it's probably not the right match, and the cloud should not try to reduce the distance between B and C at the next iteration (and concentrate on better matching parts).
Re: Fine registration ICP Normals
Posted: Mon Mar 07, 2022 8:45 pm
by Lentep
daniel wrote: ↑Mon Mar 07, 2022 1:52 pm
But actually, since ICP can potentially change the pairing at each iteration (since it pairs the nearest points, but this changes over time), this means that CC will consider that if Model B (1, 1, 1) and Data C (0.5, 0.5, 0.5) are matched, then it's probably not the right match, and the cloud should not try to reduce the distance between B and C at the next iteration (and concentrate on better matching parts).
I understand, thank you for explaining :)