Import Laserscan data -registration lost

[andysfd]

Hello,

i tried the new import laserscan function by importing a E57 file containing 54 Laserscans, exportet from Leica Register 360 (RTC360 data)

I have a problem when finishing the import:

Warning: disabled 54/54 reference records with zero GPS coordinates

The registration is lost.

[Alexey Pasumansky]

Hello andysfd,

Do you see the imported XYZ coordinates for the cameras in the Reference pane? If they correspond to the actual measured coordinates, check them on in the pane and press Update button (make sure that the coordinate system is properly selected in the Reference pane settings dialog).

[andysfd]

Thanks, one step further...

i see the camera positions and corresponding coordinates + orientation for every scan

I toggle / activate both, the position and orientation check box, when i update i get an error:

"Not enough reference data" ...
Checked the correct coordinate system

[Dieter]

Same problem here.

I first have to allign the 360°-Pictures, then it works.

But should it work without allignment? The coordinates and orientations are known...

[andysfd]

Thanks for jumping on board Dieter, i am thinking the same!

The whole Import Laserscan function is useless when it can't utilize the known pre registration. The imported Laserscans are registerted with survey accuracy (sub <1cm) and i don't want to mess this registration up.

Any idea Alexey?

[flyzk]

Following !
I need the same feature: using the accuracy of laser scan to be use as reference for my picture alignement.

[Alexey Pasumansky]

Hello all,

It would be helpful, if anyone can send 3-5 overlapping laser scans in E57 format (with coordinate information) to support@agisoft.com, so that we can reproduce the same problem and see, how the laser scans support can be improved, when the coordinate data is available.

[Dieter]

Dear Metashape team, I am happy to provide my data.

I uploaded 5 point clouds, once as an aligned Metashape project and once as original E57 files.

As you can see, the point clouds are not scanned in color. In this case it would be nice if the intensity images were not only displayed in the image processing window, but also as an image preview. At the moment, the image shown as a preview can hardly be seen.

Scan 5 could not be aligned over the 360 ​​° images because it has too little overlap. However, it has the correct coordinates and the correct rotation from our scan calculation program.

It would be very desirable if Scan 5 could now be included in the mesh processing due to its known orientation.

It would be a great advantage for us if the scans were also included in further processing without alignment, as these are normally already available with their orientation and are condensed with photogrammetry or a texture is assigned to them.

Here is the link to download: https://www.dropbox.com/sh/hsdh7dto1x3u48u/AADPS0lCShNQxg2i0nvElWVYa?dl=0

[jedfrechette]

Thanks for putting together the example data set Dieter and I agree with the other posts that the laser scan support as currently implemented is not very useful.

I think the current approach of attempting to find and match key points in a panorama representation of a laser scan is also fundamentally limited. Compared to a photo, even high resolution scans are fairly low resolution making the task difficult. Furthermore, many scans are done without color so key points that are identified in IR reflectance may not correspond to ones in color images. If scans are captured with color, that means an additional calibration between the scanner's internal camera and it's measured 3D points, which may introduce an additional source of error.

I think it would be much better to follow an approach similar to what is described in the paper:

A Fast and Robust Framework for Semiautomatic and Automatic Registration of Photographs to 3D Geometry

which describes an algorithm for using known 3D geometry to refine the estimated external and internal camera parameters. This would be a much more general solution and would eliminate many of the limitations of the current implementation.

[Alexey Pasumansky]

Dear Metashape team, I am happy to provide my data.

I uploaded 5 point clouds, once as an aligned Metashape project and once as original E57 files.

As you can see, the point clouds are not scanned in color. In this case it would be nice if the intensity images were not only displayed in the image processing window, but also as an image preview. At the moment, the image shown as a preview can hardly be seen.

Scan 5 could not be aligned over the 360 ​​° images because it has too little overlap. However, it has the correct coordinates and the correct rotation from our scan calculation program.

It would be very desirable if Scan 5 could now be included in the mesh processing due to its known orientation.

It would be a great advantage for us if the scans were also included in further processing without alignment, as these are normally already available with their orientation and are condensed with photogrammetry or a texture is assigned to them.

Here is the link to download: https://www.dropbox.com/sh/hsdh7dto1x3u48u/AADPS0lCShNQxg2i0nvElWVYa?dl=0

Hello Dieter,

Thank you for sharing the sample data.

Here is the script that applies exterior orientation to the laser scans (based on the information in the Reference pane), performs quite matching (in order to estimate the bounding box), preserving the original EO information.

Code: [Select]

import Metashape
chunk = Metashape.app.document.chunk
crs = chunk.crs
origin = chunk.cameras[0].reference.location
chunk.transform.matrix = Metashape.Matrix.Translation(origin)

for camera in chunk.cameras:
pos = camera.reference.location
m = crs.localframe(pos)
if camera.sensor.antenna.rotation_ref:
antenna_rot = camera.sensor.antenna.rotation_ref
else:
antenna_rot = Metashape.Vector([0,0,0])
rot = Metashape.utils.ypr2mat(camera.reference.rotation) * Metashape.utils.ypr2mat(antenna_rot).t() * Metashape.Matrix().Diag([1, -1, -1])
R = Metashape.Matrix().Translation(pos) * Metashape.Matrix().Rotation(m.rotation().t() * rot)
camera.transform = chunk.transform.matrix.inv() * R
camera.reference.location_enabled = True
camera.reference.rotation_enabled = True


chunk.matchPhotos(downscale=4)
chunk.triangulatePoints()
chunk.resetRegion()
After the script execution you should be able to run dense cloud or mesh generation operation based on the depth maps.

Before running the script, create a new chunk and add the laser scans (as TIFFs, if you already have them pre-saved). Then just start this script.

[Dieter]

I've done a few tests now, what can I say, it's going perfectly for me so far.

Many thanks for the very quick implementation of the function.

 

[Dieter]

Okay, I spoke too soon.....

When combining a laser scan project chunk with a photo project chunk the error message "incompatible keypoints" appears, but the chunks are merged.

Is this a problem?

[Dieter]

I'm afraid that after merging the scans with the photos, the reference will be destroyed. It looks as if the scan chunk is too large by an incorrect scaling factor, and as a result the coordinates of the chunks no longer match.

Should I make my model available or can you understand this with your own data?

[Alexey Pasumansky]

Hello Dieter,

Yes, it would be helpful to share the project (in the state prior to the merging operation) and brief description that we should follow to reproduce the problem on our side.

[Dieter]

There are two chunks: scan and photo.

In the first chunk, the scans were read in and referenced with your script. The scans were then set as a reference with an accuracy of 1 mm. Then control points for the photo chunk were determined in the scans and output as a file.

In the second step, the photos were read in and aligned. Then the photos were referenced and balanced using the control points that were determined in the scans. The control points were set as reference with an accuracy of 5mm.

In the third step, I wanted to merge the two chunks, as they are created differently (photo and scan), but are based on the same coordinates. After that I would have the dense point cloud, mesh, etc. generated.

But that doesn't work that way. After merging the chunks, the references no longer fit and I get differences of up to a few meters.


Here is the link: https://www.dropbox.com/sh/braxt55qajlcipp/AAA1LWPdZfJgC70O21tS3oEAa?dl=0