ICP did not work in

[mino]

ICP did not work in

I've started the 14days trial of Zephyr and the trial of "#A07. Laser scan handling" by my datasets.
However, the ICP did not work in one of the two datasets.
I'll send the pre- and post-ICP data of not work in to support@3dflow.net so I hope you can let me know what's going on, what's causing it, and if there's a remedy, etc.

pre-ICP data : 20200909-spray-bottle-B1-50-50 +03 step2 d5.zep
post-ICP data : 20200909-spray-bottle-B1-50-50 +03 step2 e5.zep

P.S.
As I've checked twice, I believe I didn't make the mistake the "Control point registration".

[Andrea Alessi]

Hi mino,

I had a look at your dataset :

1. yes, you correctly used the control point rough registration

2. you are having issues because unfortunately you don't have a lot of points in your point cloud, and the great amount of points in the photogrammetry scene is making the default parameters too "large" making it fail. Either decrease a lot the parameters used by the ICP, or, if you want to use the automatic x84 function, first remove the unwanted points from the photogrammetry scene

I simply used the editing tools in zephyr to remove most of the non-bottle points (the table and the support) - that was enough to get the ICP to work

I hope this helps!

[mino]

Hi, Andrea

Knowing that you simply used the editing tool and ICP worked, I tried to imitate it, and I can confirm that ICP works properly, Thank you very much!
(I realized the non-bottle points got in the way.?

I think there are three ways to get rid of the non-bottle points or prevent the non-bottle points from being generated, as far as I know, which one is best?
Or are there better ways to do than the three methods below?

(1) Narrow the boundary box as much as possible at the point after Sparse Point Cloud Generation and then enter the step of Dense Point Cloud Generation.
https://www.3dflow.net/technology/do...-bounding-box/ ...

(2) Use a masking tool.

https://www.3dflow.net/technology/documents/3df-zephyr-tutorials/tutorial-use-masking-tool/

(3) use the editing tools

I did (3) above today.
Specifically, I removed as many of the non-bottle points from the Dense Point Clouds as possible, using Tools-Selections-Manual selection, etc.

The reason I'm asking is that while Point Clouds Resistration has made the outline much nicer, some of the fine text has become distorted and blurry (compared to the results without Point Clouds Resistration).
(Please see attached screen shots.)

[Andrea Alessi]

Hi mino,

correct:

- Try to set a proper bounding box. Note that changing the bounding box size also affect the discretization. So generally you want a BB that encloses your subject only. No need to be super accurate here.
- Masking tool also has the advantage to make processing faster
- Editing tools are manual but still a valid option

If you're getting a bad texture then it seems like your ICP did not go very well. Which parameters did you use? Personally I simply cut out the excess points and run the ICP in automated mode, with the x84 statistical removal options enabled.

This is my result



I will send you privately also a link to download the zep project I made so you can double check

[mino]

Thank you, Andrea

Please send me a link to download the zep project you made.
I really want to double check!

For your information, I attached two screen shots;
(1) Selection screen for the ICP process as of yesterday.
(2) The result of today when the automatic point removal (x84) was chosen.

The latter (2) seems inferior to your results. (although it is a big improvement over my results yesterday)
The surface appears to be wrinkled. (I wonder if I didn't remove enough of the extra point cloud.)

[Andrea Alessi]

Hi Mino,

I already sent it yesterday privately / check your private messages in the forum

[mino]

Thank you, Andrea

I found your zep file in my message center.

I'm new to Zephyr, including forum, so I couldn't understand what you were saying. (My English ability is also weak) Please forgive me.

When I looked at your zep file, it looked almost the same as my retry result that I chose Automatic point removal (x84) and then applied ICP.
(Your results also appear to be wrinkled when the "Enable/Disable Light" swtch are turned off, similar to my retry result.)


Today, I tried ICP on a different subject from the previous day, and I was disappointed that there seemed no difference between the results with and without Automatic point removal (x84).
(Selecting Automatic point removal (x84) does not necessarily shift the results in a good direction..)

[Andrea Alessi]

Hi mino,

no worries!

my screenshot was meant to show that a correct ICP will create a good texture, while your seemed skewed a bit due to probably a slight misalignment.

The mesh quality ultimately depends on the quality of the imported point cloud, so I would probably simply get a more dense/cleaner scan if you require a better geometry

In any case I'm glad to hear that you managed to do the ICP correctly