Advice : which type of distance should I use?

[ZAIDAN]

Hello,

I am doing tests, using a 3D scanner laser to measure the erosion of the soil in a flume . The maximum erosion that I can obtain is 6 cm. I want to minimize the error at the maximum, and in order to be able to compare my results, I want to represent all my results in a plane 2D (z=0), with the same number of points.
I am new at cloudcompare. So, I want to know which type for computing distances in cloudcompare (C2C,M2C,M3C2..) is the best in my case and you recommend me to use it? and what will be the maximum error that I will obtain ? Can I represent using cloudcompare, the error that I obtained for each test?

Thanks in advance,
Jana

[daniel]

For maximum accuracy, I would advise to use M3C2: https://www.cloudcompare.org/doc/wiki/i ... 2_(plugin)

You will have to dig in the theory (article, some posts on this forum, etc.) to fully master it.

And then yes, you can definitely retrieve the maximum error (from the resulting distance scalar field) as well as other things.

[ZAIDAN]

Hello,

Thank you for your response.

I try to use the M3C2 method and it seems good.

But, I have some questions :
1) I noticed that the density of points decreases when the distance between the scanner laser and the object scanned increases.
This change in density doesn't influence my measurement ?! For core points far from the scanner, the number of points averaged to determine i1 and I2 will be different from those near the scanner ! That why I have doubt in the choice of the projection parameter "d".
In your opinion, how can I choose the value of "d" in order to be best for my measurements ?
2) Is the registration error mean the precision of my scanner laser used ?!
3) I am using Scanner photoneo size S, can I represent with this instrument the precision maps ?!

Best regards,
Jana ZAIDAN

[daniel]

1) Well as long as you have a sufficient number of points in your 'local neighborhoods' (i.e. within the 'scale' radius), then the statistics should remain reliable. Of course this means that you'll potentially have a lot of points in the areas closer to the scanner (which will mostly slow down the computation).

Another option is to subsample the cloud spatially so as to get a more even sampling/density. But I think it's not necessary.

2) It's more the precision of the registration process, if you had to register one cloud with respect to the other (or to a common reference)

3) Precision maps are generated by Photogrammetry tools only I believe

[ZAIDAN]

Thank you for your response.

I try to compute the density (N neighbors) at each point in my reference cloud and the compared cloud for R=10.
For the reference cloud, I got a minimum value of 17 located in the region far from my scanner and a maximum value of 5718 located in the region near my scanner.
For the compared cloud, I got a minimum value of 80 located in the region far from my scanner and a maximum value of 6115 located in the region near my scanner.
In your opinion, taking a value of d equal to 2*R = 20 will be oK to apply the M3C2 in this case ?! Are there a risk of degradation of my resolution far from my scanner ?!

On the other hand, how can I proceed to estimate the value of the regitration error in my study ?!

Best Regards,
Jana ZAIDAN

[daniel]

Well that looks fine (you could probably even use R only).

And for the registration error, either you know it (thanks to registration spheres, or something like that), or you can use the 'Align' tool on equivalent points (but you'd better get a lot of them so as to minimize the approximation error).