Close-Range Photogrammetry with Local Coordinates doubts

[Frmoraes]

Hi,

I suspect that my problem may exist only because of my lack of knowledge, any guidance and tips would be highly appreciated.

I am a PhD candidate and I've been doing some research in Close Range Photogrammetry (CRP) to, in a near future, analyze deformation in concrete beams.

I am trying to create a model to analyze the RMSE of some markers, that I’m using as checkpoints. I have taken 84 photos, with 80% overlap (horizontal and vertical), at a distance of almost 1 meter from the object (as it can be seen in the PHOTO1 and PHOTO2).

I have been using a Canon EOS R, so I don’t have any coordinates from the photos. Nonetheless, I created 77 markers inside the object that I’ve determined the coordinates with a Total Station (TS) to orient the model (set in X = 1000; Y = 100; Z = 5000). In my experiment, I would use up to 50 control points and 27 as checkpoints.

My workflow is:
Create a project;
Add the 84 photos;
Align the photos (Accuracy = High; Generic preselection = True; Reset current Alignment = True; Key point = 40.000; tie points = 10.000; Exclude stationary tie points = True);
Detect Markers (all the 77 are detected );
Add the Coordinates;
Update Transformation;
Optimize Cameras;
Export Report.

My problem is that after “Add the Coordinates”. Until this point, the model is shown correctly just like in the real world (PHOTO 3 is angled to show the ambient). However, when I load the txt file with the coordinates and Update the model, the cloud points come out in a different way (Photo 4) and the RMSE report showed a high error in Z_axis when compared with the other two axes (Photo 5).

I’ve been trying some solutions and I released that for some reason the Z_axis has crescent values when pointing out from my monitor (too myself) and I wanted that its values were crescent when entering in the monitor.
As I’ve read in the forum that I can’t change the axes configuration, as I want in this problem. So, I’ve tried two solutions. In the first one, I’ve changed coordinates from axis X to axis Z, and in the second one, I’ve changed all coordinates from Z_axis (5000 – (Z_axis – 5000)). The RMSE reports (PHOTO6) were way better than before and they were similar, just changing Z_axis with X_axis.

Should I use one of my solutions? If I should, which one? If not, what should I do? Is there any better option?
Any pointer would be appreciated.

Ps.: If I had to choose, I will use the first solution, because I don’t have to modify any coordinates. I’ll only need to generate reports with ZY axes instead XY.

[Kiesel]

Hi Frmoraes,

I'm surprised that you use such unstable material as cardboard for such task, you will not get good results with it.

The coordinate system you get with a total station differs to that in your pictures, where normally z is height and not y. Perhaps that's causes the problems you have after importing coordinates.  How is your project coordinate system defined? Could you show the Reference settings window?

Exclude stationary tie points is for turntable tasks, which you don't have.

For such task, analyze deformation of concrete beams, it is important how the pictures are taken. I can recommend to take pictures like you do for camera calibration, so add +/-90° and 180° rotated and oblique photos to get a stable camera configuration and good camera calibration.

Best regards,

Kiesel

[Frmoraes]

Thanks for the reply, Kiesel!

About the cardboard, I'll change this scenario. In my new test, I'm going to put the markers in the body of the object (that is wood).

The coordinate system I've got with a total station was set to X=1000; Y=5000; Z=100. And as you said Z is the height. For the Metashape, I firstly changed Z_axis with Y_axis, and then I got photo4 as result.

In one of my future tests, I'm going to add oblique pictures (something like a Maltese Cross configuration) and compare the results with only "central" pictures.

When you said about adding +/-90° and 180° rotated pictures, I should take just some pictures in this configuration, or I should take again all the pictures in this configuration?

My "Reference Settings" is shown in "RSettings.JPG" and "ImportCSV".

Thanks,

Frmoraes

[Kiesel]

Hi Frmoraes,

About the cardboard, I'll change this scenario. In my new test, I'm going to put the markers on the body of the object (that is wood).

That's better!

The coordinate system I've got with a total station was set to X=1000; Y=5000; Z=100. And as you said Z is the height. For the Metashape, I firstly changed Z_axis with Y_axis, and then I got photo4 as result.

I wouldn't do this, because it isn't necessary. Also your wooden plate isn't standing vertical as can be seen in photo 2. You can later easily define an (orthogonal) orthomosaic and DEM to markers at the wooden plate.

In one of my future tests, I'm going to add oblique pictures (something like a Maltese Cross configuration) and compare the results with only "central" pictures.

Good!

When you said about adding +/-90° and 180° rotated pictures, I should take just some pictures in this configuration, or I should take again all the pictures in this configuration?

Only some are enough.

Also you should tick the local coordinate system for markers reference in reference settings.



Kiesel

[Frmoraes]

Thanks Kiesel

In my next experiment, I'm gonna try adding rotated pictures.

About the coordinates, I've used the normal XYZ (as it is used in Total Station) and on the report I just had to select "front XZ", nothing so difficult.

Thanks again!!!

[Kiesel]

Hi Frmoraes,

yes and don't forget to shoot oblique photos too, the better the photo intersection the better the bundle optimisation.
If I were you, I would investigate in what is the best camera configuration and have a look at camera calibration too.

Best regards and success,


Kiesel