Why does this componet goes into the ground? +RTK +GCP

Development Asset Creation
, ,
1

Good afternoon community, I´ve come here seeking for help. I hope I can provide enough information so you could help me, I thank you in advance.

What I´m doing is genrating a model of a infraestructure from pictures taken from a drone (Mavic 3 Thermal) with the RTK option activated, so the georeference should be pretty precise.

I´m going to try to define everything I´ve tried so far:

It was a cloudy day, and the lightning was constantly changing, I was trying to compensate with camera parameter, but there are some photos with more lighting than others. This was a problem at first with the allignment, but I started placing control points and managed to merge a lot of the different components (I´m focusing on the roof of the infraestructure).

1
11920×1026 361 KB

What I did was placing control points and doing the “merge components” action, this gave me almost all the roof covered, but what I´ve encountered is that two of the components, are warping into the ground (they should be completely flat).

2
21920×1026 220 KB

What this two components have in common is that they are only composed of 90º ortho images from the dron (even though there are some other pictures with another inclination).

What I´ve tried to do is:

-Tweaking the alligment setting trying to increase the detection and lowering the expected overlap:

4
42547×1162 114 KB

-Placing control points with this components, but I get the error “the projection error of the control point is over the limit”, and I dont understand this, because I place the CP exactly in the same place in multiple photos. After placing the CP I allign again, but getting the same result.

-Geo-merging components (I though I would get awesome results since I was flying with the RTK activated, but the result is pretty bad).

What I´m thinking I could do (would love to hear your suggestions:

-Delete the images that are giving me problems, alligning everything again and then try to work it out from there (I don´t even know if I can delete certain photos).

-Trying to see if I can merge this bad components with the other images of another camera yaw, and once they have this other angle information, they may work (?)

As you can see, this is a huge project and had to submit a lot of drone permits to do it, so I´m pretty frustrated with this, I would be so amazed if you could help me

3
31920×1024 299 KB

Thank you so much in advanced, Im craving to hear your opinions.

2

Hello Johny,
it looks like there is something wrong with your dataset, as only small part of all images is aligned into one component.
Have you followed all basic photogrammetry rules for taking images (overlap, coverage, capturing path)?
Also, how sure are you with the RTK coordinates? Have you validated those before? If pretty sure, then you can try to set better precision for the prior poses in the alignment settings.
You mentioned you used GCPs, but from my point of view it looks like you are using only control points. Or do your GCPs have measured coordinates?
On your place I would start over, importing small parts of the images into RealityCapture, align, add another part of the images, align/merge etc.

It is not possible to delete the images from the project after their import. Therefore it would be better to start over.

Also, you can try to turn of the camera priors and check, if the alignment will be better after that (as you have a pretty big position error in your aligned component).

3

Hello Otrhan,

Thank you very much for your response, I´ll reply to what you ask:

Yes, I followed the photogrammetry rules the best I could; the roof was my main objective, but I additionally captured the walls, which didn´t turn out pretty good in the model, so now I´m focusing in only generating the roof model.

I´m not pretty sure about the RTK coordinates, haven´t validated them.

What do you mean by “you can try to set better precision for the prior poses in the alignment settings”?

I didn´t mean to place “GCP”, as I didn´t have them, what I´m doing now in setting CPs in matching points of images where the program struggles to generate or merge the components.

Could you please elaborate more about " you can try to turn of the camera priors and check, if the alignment will be better after that (as you have a pretty big position error in your aligned component)."?

I will follow your advice and start over again with small parts of images and merging them.

Thank you very much for your help, it really is appreciated.

4

Hi Johny
Regarding the better precision:
In the alignment settings you have an option to set the position’s accuracy. As you used RTK drone, it should be better than predefined 10 meters:

image
image427×397 11.6 KB

You should set it the the drone’s clarified precisions.

Regards the camera priors, you can find the settings under ALIGNMENT/Registration/Settings/Camera prior settings (check the previous image) and you can set it to No. But then your model won’t be georeferenced and scaled.

Another option is select some images which had the best accuracy (maybe the ones capturing the roof from the above) in 1Ds view, revert the selection and set their prior pose to Unknown:
image
In that case, good images will keep the scale and georeferencing, and bad images will take this information after their alignment.

Good luck with your work!

5

Greetings Otrhan,

Once again, thank you so much for your help, I really appreciate it; I haven´t responded in over a month because I cant give full priority to this job, and it is starting to get on my nerves…

I tried fixing the position accuracy to the drone specs and it really helped, I almost had all the roof, but there are some components that doesn´t merge, and Im talking that I even have multiple components of 2-5 images that have plenty of overlap and that even a 5 years old could point the tie points, but RC refuses to merge them, even after placing control points.

I then tried to start from scrath following your instructions, but it is even worse, and I cannot understand why; for example:

Photo capture (can´t upload more than one photo on the post)

I tried alligning this segment shown in the screenshot, in which I moved the drone horizontally, from left to right capturing images, which should of course, give me the straight wall, insted I get this!

Screenshot_4
Screenshot_41920×1150 183 KB

A complete circle…

I don´t know what else I can do, this project is driving me crazy, and I already paid a digital twin software in which I would represent the RC exports, and I can´t even get it…

Do you know of a professional that I could pay to get this done? I would really love to get it done myself as I have always done, but I´m encountering too many problems…

Once again, thank you.

6

HI Johny, I am sorry to hear that.
Regarding the shown issue, it is the general photogrammetry problem called a banana effect: Tutorial: Banana Effect - What To Do If My Model is Bent
Basically, one line is not good enough to align the image properly. To solve it properly, that part should be captured at least in a grid.
I suppose you will need to have bigger image groups to align all data properly.

Regarding to the merging component, what is the capture angle between those components? Is it under 30 degrees? How do those images with placed control points look?

I think you will be able to find someone in the Facebook CapturingReality Arena’s page: Redirecting...

7

Hello Otrhan,

Really interesting about the Banana Effect, It´s not the first time it happens to me, and I now know what that is.

Following what you just said, I will try to allign all of the building, not only the roof, with the previous instructions you gave me, so maybe it helps with the banana effect and then I can only keep the roof.

Regarding the merging component, the angles doesn´t vary between them… So I don´t understand the problem.

I will keep this updated.

Thanks again.