Noise reduction in point cloud

[jfhullo]

Hi all,

Since CC has a function to compute density, could we imagine a noise reduction function of the point cloud. It would then offer a better with EDL by deleting lonely points.

Is it already possible using a tric or will it require a new functionnality?

Thanks in advance.


JF

[daniel]

Indeed, you can do it in two ways:
- remove noise by computing "roughness" on the point cloud and then keeping only the points with the lowest roughness
- remove the isolated points or groups of points by computing the "connected components" (with a rather low octree leve)

However, a dedicated algorithm would be a plus for sure.

Daniel

[Ingsayyad]

How can I filter human or cars from cloud. Any trick to avoid manual segmentation.

[jason3d]

You can look at the cloud from the bottom and use hidden point removal, I have found that useful to get rid of vehicle stripes in roadways.

[Dimitri]

Jason3d, you've just enlightened my day !
I've never tried HPR before, and it's just great !

How can I have missed this for the last 5 years I've been using CC :-(

[daniel]

I must admit that it's a very clever way to use HPR ;)

[dmbrubac]

You can look at the cloud from the bottom and use hidden point removal, I have found that useful to get rid of vehicle stripes in roadways.

Sorry for the really newbie question on this - I only started using CC yesterday. I'm also new to working with point clouds in general.
I have a parking lot full of trucks and I want a cloud just representing the parking lot, with or without holes where the trucks used to be.
I segmented out a group of trucks to try this method and was able to view the cloud from the bottom and switch to perspective mode so HPR would run.

It seems like the only way I could remove hidden points was to tip the cloud so the pavement hid the trucks, more or less. This took out some of the points, so I used the new cloud, tipped it a different way and got rid of some more. Even after 5 iterations I still had bits of tires and bumpers but my parking lot was also looking pretty ragged. Am I missing something?

It seems a method might be to draw a polygon around an object and then delete points based on a height histogram or something similar. Ideally then you could remove noise and interpolate a patch across the hole.

I'm also looking at CANUPO to classify pavement and not pavement, but I haven't got my head around how to do that yet.

If it makes a difference, my clouds are generated out of Pix4D from photos, not lidar.

Thanks for any input.

[daniel]

I'm not sure about the HPR-based method as I never tried it myslef.

But indeed you can use the scissors tool (http://www.cloudcompare.org/doc/wiki/in ... ation_Tool) to cut out the points inside a polygon.

And you can also convert the height to a scalar field to segment the cloud based on the height (with http://www.cloudcompare.org/doc/wiki/in ... )_to_SF(s) and http://www.cloudcompare.org/doc/wiki/in ... r_by_Value).

And for Canupo, it's a bit longer but don't hesitate to look at the wiki (the idea would be to make a class with the points which are flat at all scales - contrarily to the others that shouldn't be flat at all scales, especially for large scales).

[intenlia]

Hi all,

I want to know how to remove the small tress in the point cloud.
I found some paper that uses roughness for removing small trees, that he removes point that has roughness values larger than 0.10.
I would like to know, how to remove the roughness less than 0.10.
I already counting the roughness. but I don't know how to remove the points that has a specific value.
Please, help me to solve this.
Thank you for your help.

[daniel]

Use the 'Edit > Scalar fields > Filter by value' tool. You can also interactively play with the scalar field display parameters (in the cloud properties).