Hi Bileh,

I would not bother with MATLAB (but please read on). There are dedicated - and FREE - tools available to do this, like CloudCompare. https://www.danielgm.net/cc/

CloudCompare is a very powerful cloud processing software with many unique tools. It also has a very fast and robust version of the ICP algorithm implemented: http://www.cloudcompare.org/doc/wiki/index.php?title=ICP

Check their YouTube channel for many helpful tutorials: https://www.youtube.com/watch?v=MQiD4HjhpAU&list=PLBNUxsUA00UAT63O0d95pByrCjtqlXN4_

However, CloudCompare is limited to an ICP between two clouds, so error accumulation can indeed occur. This is solved by OPALS, another extremely powerful software for point cloud processing which is programmed at the TU Vienna: https://www.geo.tuwien.ac.at/opals/html/index.html

In contrast to CloudCompare, OPALS can work with various clouds at the same time. For more information, see: https://www.geo.tuwien.ac.at/opals/html/ModuleICP.html

The above procedure of OPALS can also be found inside MATLAB. Philip Glira - of the TU Vienna and one of the developers of OPALS - programmed a point cloud toolbox for MATLAB which also includes this co-registration (with global error distribution) of many point clouds: https://www.mathworks.com/matlabcentral/fileexchange/54412-point-cloud-tools-for-matlab

You can see a graphical representation of the global error minimization here: https://www.geo.tuwien.ac.at/downloads/pg/pctools/pctools.html

Finally, there is also the Open3D toolkit: http://www.open3d.org

It includes a so-called multiway registration to co-register multiple point clouds: http://open3d.org/docs/tutorial/Advanced/multiway_registration.html

I hope this helps you further.

Cheers,

Geert

Hi Mr. J Verhoeven,

thank you for your detailed response.

here you will find an image of my 3D reconstruction with ICP to get a better idea of what I'm doing.

Thanks,

Bilel

Hi Bilel,

interesting. Did you try any of the mentioned packages?

Cheers,

Geert

Hi Herr J Vorhoeven,

i took your recommendations into consideration and tested the multiway registration of the open3d library and the results are not good (pc_3_als_output).

The are 38 point Clouds and the problem is always, how to tie the first and the last(it ist the first after a rotation of 360°) point clouds and distribute the error

I use MATLAB as part of my Master project and I have to explain with simple calculation examples how the algorithms work.

For global registration, I am programming "Registration using the Helix Transform" of Nüchter's article "Linearization of Rotations for Globally Consistent n-Scan Matching". I think it's the most suitable algorithm for my case.

I look forward to your reply

Cheers,

Bilel

Hi Bilal,

well, I am not an expert on the core of these algorithms. I more or less listed what is available. If you would like more in depth info, I would contact Philip Glira (who made the above-mentioned MATLAB toolbox). Did you try that one?

Good luck!

Geert