Point clouds in Autocad 2016

[nadar]

I will also continue testing on my side and keep you informed.

First test with another project of a single facade using Dieter's workflow. The main plane is perfectly detected, but secondary planes (such as doors, etc). are missing. I can live with that however.
I wil do more tests on complete buildings (4 walls, roof, etc.)

[Dieter]

The following results:

1) point cloud scan exported with PTX, imported with recap, segmented and output as RCP.

Many surfaces are automatically detected and are right in the spatial representation.

2) The same point cloud scan exported to ASCII, imported with recap, segmented and output as RCS.

Only the largest surfaces are automatically recognized. Tapping the edges these surfaces are shown only partially accurate.

To test this further I need an ASCII PTX Converter, someone knows there a solution?

With my SCENE program from Faro, it dont work, as it reads the ASCII format incorrect.

[James]

PTX files include the scanner location for each scan, which is useful for calculating normals as you can always tell which side of a surface the points refer to etc.

You could import PTX to recap, and then export from recap as PTS, which gives a plain text file including only the position of points, and no 'normal' information, if that is what you want to do?

PTX is also plain text, but each point is stored relative to the position of each scan setup, so have to do some matrix calculation to convert from local scanner coordinates to world coordinates.

[Dieter]

We want to find out why Autocad Recap finds surfaces in some scans, and whether the file format plays a role here.

[James]

What do you want the ascii ptx converter to do? i had possibly misunderstood that you wanted to convert ptx to ascii (which is fairly trivial, assuming by ascii that you just mean a plain text list of 3d point coordinates?)

Conversion the other way - plain text list of coordinates to ptx - is not possible unless you know the location from which each point in your ascii file was scanned.

in photoscan this seems a bit unlikely as each point in the cloud is the result of multiple cameras all at different locations.

[Dieter]

This all sounds very plausible. That could also explain why recap takes only a few surfaces, if the original data was generated from pictures. It lacks the information of where the point has been recorded.

I thought that would be described by the point normals.

[nadar]

same results, using a .LAS file generated by Photoscan.
I tried several options regarding cloud decimation, but no significant improvement: Recap recognises some planes, but some are correctly located, other are slightly mis aligned and many are completely wrong.
A pity we cannot select manually the planes we want to use as reference (selecting 3 coplanar points, for instance).
In its current form, working with "photogrametric" clouds (i.e. without known scan stations), the recognition of structure in the cloud is nearly useless.

Is there another solution to generate good reference planes ?
Placing a few UCS is not efficient because I can't snap the UCS axis on the cloud.
Is there a solution to generate a volume precisely matching a part of a cloud (e.g. a wall) and use it as a reference to define a new UCS.
As you can see, I'm a pure newbie in Acad, but I thing this software has great potential
 

[nadar]

What is the difference between .RCS and .RCP files ? If I don't use the possibility of using structurated planes, I don't see any difference, and generating .RCP seems faster. Could somebody please confirm ?