tank analysis with ransac problem

[sewro]

Hi, I'm trying to do deformation analysis of a tank, I imported a pts point cloud, created a theoretical cylinder. The 2D map shows maximum values of 0.45m but in the generated dxf files the values go up to -2.71m for example. I set the drawing magnification when generating the dxf to 10. I tried different ways and each time the values are similar. Why is there such a discrepancy between the dxf files and the generated map and scale. I attach a 2D map photo and a screen shot of dxf

[daniel]

Hum, hard to debug like this.

Could you share the data with me? (so that I can recreate the map and export it to DXF). You can send an email to admin@cloudcompare.org

[sewro]

Sure, I sent the point cloud via email.

[user3]

Hello,

I’d like to use RANSAC for outlier removal in a point cloud. The idea is for each random subset of points to calculate the least squared plane and check the whole point cloud for outliers (within the given distance threshold or MAD). This would go iteratively until the max number of iterations are fulfilled. The Automatic RANSAC plug in fits shapes and it’s rather unsuitable for this task. However, the roughness (from Geometric Features) is the distance between the point and the best fitting plane for each point. Is it possible in CC to remove outliers by running best fitting plane on the whole point cloud without shape fitting? Can you please indicate where this is in the code?

Thank you very much in advance.

[daniel]

This really looks like the Noise filter algorithm (the only difference being that the SOR algorithm will go through all points, one by one, instead of being random).

See https://www.cloudcompare.org/doc/wiki/i ... ise_filter

[user3]

Thanks for the hint! Apparently RANSAC doesn't perform good for outlier removal in 3D point clouds.

Another thing, can a best fitting curve or B-spline (using least squared plane or RANSAC) be calculated in CC?

[daniel]

Ah no, that's not possible at the moment.

[user3]

I've coded something, however the cubic, quadriatic and polynomial curve (red) don't follow the object's shape. It's supposed to fit similarly like the green line. Do you maybe have an idea how to approach this?

[daniel]

Have you tested the 'contour extraction' tool? https://www.cloudcompare.org/doc/wiki/i ... ngle_slice

If properly parametrized, it can extract the 'concave' hull of a point cloud.