qCANUPO (classifier files, etc.)

[daniel]

Hum, no sadly, I fear it depends on the 'scales' of the classifier, and on the borders, the classifier is missing some information required for the larger scales).

[lightcuber2]

Hi @PablerasBCN,
The edges of point clouds are a problem for geometric features of planar objects, since the principal components on the edges are numerically different from their values in the middle. Half the sphere is empty on the edge so the principal components are skewed.

You could try including some parts of the edge of the cloud in your training data. That would tell the classifier that you want edges to be a particular class.

But it might then classify all edges the same way, depending on the characteristics of your different classes. It might be worth a shot. In general I wouldn't recommend this strategy because of this problem of edges all looking the same, but it just popped into my head. :)

If you are classifying vegetation vs ground for example, I think something like this could work.

[Dimitri]

Dear all,

our team has released 3DMASC, a version of CANUPO on steroids :
. you can have any number of classes
. you have a much larger range of geometric features you can use, but also intensity, RGB, Echo based features (for Airborne LiDAR) (CANUPO was only using 2 geometric features per scales)
. for advanced users, you can even use 2 different point clouds (2 different wavelengths, or 2 points clouds at different time)

Because 3DMASC is much more advanced than CANUPO, it's a little bit more complex to use, but the benefits in terms of classification accuracy and flexibility are really worth ! CANUPO can still be useful for teaching, but if you are using Canupo for classification of your point clouds, I strongly suggest you move to 3DMASC.

The description of the plugin can be found here:
https://lidar.univ-rennes.fr/en/cloudcompare

The submitted paper describing the plugin used in the context of bi-spectral topo-bathymetric LiDAR is here:
https://hal.science/hal-04072068

We'll add progressively video tutorials for TLS, SFM and ALS data.

Enjoy, and let us know if you find any bug in the dedicated 3DMASC section.

[PablerasBCN]

Dear all,

our team has released 3DMASC, a version of CANUPO on steroids :
. you can have any number of classes
. you have a much larger range of geometric features you can use, but also intensity, RGB, Echo based features (for Airborne LiDAR) (CANUPO was only using 2 geometric features per scales)
. for advanced users, you can even use 2 different point clouds (2 different wavelengths, or 2 points clouds at different time)

Because 3DMASC is much more advanced than CANUPO, it's a little bit more complex to use, but the benefits in terms of classification accuracy and flexibility are really worth ! CANUPO can still be useful for teaching, but if you are using Canupo for classification of your point clouds, I strongly suggest you move to 3DMASC.

The description of the plugin can be found here:
https://lidar.univ-rennes.fr/en/cloudcompare

The submitted paper describing the plugin used in the context of bi-spectral topo-bathymetric LiDAR is here:
https://hal.science/hal-04072068

We'll add progressively video tutorials for TLS, SFM and ALS data.

Enjoy, and let us know if you find any bug in the dedicated 3DMASC section.

Somehow this event passed under my radar!!

much excited!!

I've been manually segmenting a road mobile mapping scan and I've some Kilometers for training and to compare against as ground truth data

thank you so much!

[daniel]

See this interesting post on Linkedin about 3DMASC: https://www.linkedin.com/pulse/automati ... bderrazzaq