City Site Modeling

[Nemooo]

Hi guys,
I'm Niccolò de Ruvo, a student of the International Course on Architectural Design in Florence. I'm doing a master's thesis on the Automated City Site Modeling using a 3D cloud point from a laser scanner. I'm in a very starting point and after reasearches I found that CC can help me with RANSAC and CANUPO. I'm working on an ancient fortress that is like a small district of a city, with green, buildings, streets and empty spaces. I have to automatically distinguish those elements. Am I on the right way? Can you suggest me something that can help me?

Thank you.

[daniel]

I fear Ransac is a bit too simple for what you are trying to achieve, even though it can give interesting results and it may be a starting point for something more evolved.

And CANUPO is used to classify natural scenes or specific features (such as powerlines, etc.). I don't know how it would perform for the purpose of segmenting architectural buildings.

There are hundreds of fancy algorithms and methods for this purpose... but I don't know if any of these have ever reached an open-source (or even commercial) application.

[william.nguatem]

Since this is my first post in the CC forum, I'll start by thanking the entire CC Team for the great job.
It would have really been such a huge pain for me without you guys. I work in a domain where we model out-door scenes from point clouds from LiDAR and SfM/MVS, thus it is normal to have 800+ Mio. 3D Points.

Now to your question:
For the task of City Modeling from point clouds, RANSAC can really be limiting. I ran into the same problems last year and finally had to devise my own algorithm. Currently am discussing with my project partners to see if I can release the code. Meanwhile, you can look at the results in the following paper. If you think this may bring you ahead towards your task, then you/we may want to give it a try ?

Paper: "Contiguous Patch Segmentation in Pointclouds"
http://link.springer.com/chapter/10.100 ... 45886-1_11

If you are well versed in stochastic processes then I can assist you to re-implement the things, whilst am waiting for my folks to agree that we release the code. Am sorry if this is a bit challenging to understand, RANSAC wasn't really working for me, hence something a bit grey.


Sure, if am allowed to release the code and CC see the need for it, then, this will really be a great opportunity.

[daniel]

I'm pretty sure a lot of people will be interested to have the ability to test your algorithm (in CloudCompare even more ;).

I shall give a presentation of CloudCompare at 3D Arch 2017 (http://www.3d-arch.org/). I'm sure it would be a great hit if we could have a running version by then!

[william.nguatem]

Daniel, thank you for your prompt response.

I'll keep you and this thread too updated as things unfold.

Regards
William

[daniel]

Is it possible for you to send me a pdf version of the article (cloudcompare [at] danielgm.net)?

Thanks,