M3C2 core points and direction of distances/normals

[uasghar]

Hi Daniel and Dimitri,

I'm using M3C2 plugin to compare a series of point clouds, and so far I've been satisfied with M3C2 results compared to C2C. I should mention that I read the corresponding publication in which M3C2 algorithm was applied to the Rangitikei canyon point clouds. However, I have a couple of questions with regards to M3C2 use.

1) For two point clouds, each with the resolution of 5cm, guess parameters were D=6.16m, d= 0.823m, L=58m and dx=0.421m. However, If I choose D= 0.5-10m with 1m interval, d=2m, L=10m (knowing that there's no feature on site that is 58m high or wide), and dx=0.4 m, am I on the safer side compared to what the software guessed?

2) For the direction of normals I used +Z, does that mean that the computed distances will also be in Z direction? If yes, is there an option with the normal direction menu to compute net distances?

3) Keeping all other parameters constant, I varied dx from 0.8, 0.7, 0.6...0.2m, however, there was no significant difference between the calculated distance from 0.5 to 0.2m. From the perspective of saving computation time without affecting the accuracy of results, does that mean dx= 0.5 or 0.4m is enough to get accurate results.

Regards,

Asghar

[daniel]

You'd better trust Dimitri on this, but here are my answers:

1) Yep, the human guess is always a better than the 'guess best params' button ;). Seriously. the plugin has no clue about your data and the actual scale of the objects you are interested in. It only make some considerations about the cloud density and the local flatness.

2) it's only the normal sign that will be changed so that the Z component of the normal will be positive (i.e. no normal vectors will point downward). What do you mean by 'net distances'?

3) Yes

[uasghar]

@Daniel thanks a lot for the reply.

in Question (2), by net distance I mean the total distance between the two points and not their X, Y, Z components. like d = sqrt(dx^2 + dy^2 + dz^2).

C2C algorithm provides four distances if "split (X, Y, Z) components" option is checked i.e. total/net distance d, and its three components dx, dy and dz.

Because M3C2 only provides one distance, I'm confused which distance is it? the M3C2 publications says that distances are computed (between the points at scale at which the local surface appears most planar) along the normal. I used Z+ for normal orientation because I used vertical aerial imagery, so are my distances dz or d?

Regards,

Umair.

[daniel]

Yes this is the real 3D distance ('d'). To get only the Z component, you would have to force the normals to be purely vertical (see the 'Calculation mode' options).

[uasghar]

@Daniel, thanks a lot for clarifying that.

There's one more question about .bin file of M3C2 distance. Today when I reopened M3C2 distance.bin file for my project, there was a shift/translation in the coordinates of the point cloud compared to the original point cloud (see fig. below). So I'm why if .bin file system alters the coordinates of the points in the cloud?

Please note that I saved the M3C2 distance point cloud as .bin by file>save option.

Regards,

Umair.

[uasghar]

Canupo Classification Figure Zoom-In example.jpg (79.9 KiB) Viewed 8301 times

[daniel]

It's probably an issue with the Global Shift information.

Can you check that the clouds have the same Global Shift information? (Edit > Edit Global Shift & Scale). You may have applied a different global shift the second time?

[uasghar]

@Daniel,

Both have the same shifts. Also, I opened both clouds at the same time by file>open and multiple slection using ctrl key. The problem persists.

Regards,

Umair.

[daniel]

Hum, that's odd. I don't know what to say. Do you have any mean to detect at which point in the process this shift occurred? Like calling M3C2 again and save the result and open it again?