A new software for classification of 3D point clouds

[Dimitri]

Hi everyone,

I'm note sure if it's the right place to post this information (Daniel, feel free to move the topic or even cancel it if you find it's innapropriate). With Nicolas Brodu, we have developped an open source software that automatically classifies 3D point clouds according to their loacl geometrical properties. One of its main application is the automatic classification of vegetation, but it is more generic than that. While it's independent of CC for operation, we use CC all the time for point cloud vizualisation as well as point cloud segmentation according to scalar. If you're interested in this software you can download it at Nicola's Brodu website :
http://nicolas.brodu.numerimoire.net/en ... index.html (linux and windows binaries, as well as source code)
and find a user guide and various tutorial on my freshly launched website :
http://www.geosciences.univ-rennes1.fr/ ... rticle1125

The reason that I've placed this in the "requests, new functionalities", is that if many people are interested in this classification software we could try to find a way for a simplified version to make its way into CC. There are 2 different aspects in the software suite :
. The classifier construction which would remain a standalone command line application because it's not really often used (once you're happy with your classifier to separate trees from ground, or grass from roads, you don't want to change it everyday).
. And the classification of point cloud according to existing classifiers. This part could make its way into CC as in that case, CC just need to estimate the multiscale dimensionality characteristics of the point cloud (it is based on a PCA done within a search radiusat various scales) and apply the probabilistic classification according to the classifier that would be given as an input file (or very successfull classifiers could stay permanently as options in the CC plugin).

Anyway, this is just in case people are interested in this kind of algorithm...

[jfhullo]

I'll have a deep look to your work, but it really seems awesome. I'll give you (or a student of my lab at INSA Strasbourg) soon.

I'm really excited by the release of your work.

Thanks for sharing!!


JF

[jfhullo]

Really nice piece of software. Can we dream of an implementation in CC Dimitri?

JF

[Scrible]

I have to agree with jfhullo. It seems as if this software has enormous potential. Having it in CC would be great!

Dimitri,
I've been using the software within my master thesis and I'd be glad to give you some feedback as soon as some other deadlines are met.
My sincere admiration for the development of this software; I'm suprised it hasn't arrived in the world of 3D laserscanning with a "bigger bang".

Cheers,
Arie

[Dimitri]

Hi,

thanks for the appreciation. It was really a team work with Nicolas Brodu bringing in his expertise in applied math and code development, and me just bringing in ideas about multiscale analysis of 3D features (and the scientific needs !).
Unfortunately, my level of programming in the context of "operational" software is quite mediocre (and I've not yet fully evolved towards C++ !). So I won't be able (not will have time) to implement a simplified version of the classifier in CC.
Nicolas is now working on another topic and won't be working on tls stuff in the foreseeable future.

But if someone is interested in trying, I'm sure Nicolas and I could help him navigating through the code and algorithm. The hard part is really the classifier construction which I think should stay apart from CC. But the calculation of the multiscale parameters and the application of the classifier in CC should be "relatively" easy to implement.

Cheers

Dim

[jfhullo]

Dear Dimitri,

I've used with a lot of interest Canupo. I would anyway suggest (if you have some time) to add elements to the tutorial, since the use of the several modules is not straightforward. But maybe this project is already over.

JF

[Dimitri]

Hi JF,

I'd be glad to improve the tutorial. Right now I'm very busy and still have to finish the tutorial for M3C2, but it'd be great to have your feedback.
Maybe we can continue this over email or phone ?

Dimitri