Forum

Author Topic: Large Error in GCPs  (Read 4321 times)

Shovelhead

  • Jr. Member
  • **
  • Posts: 75
    • View Profile
Large Error in GCPs
« on: February 13, 2018, 12:19:05 AM »
Im working on a thesis project for my bachelors in Geomatics where I compare the accuracy and suitability of UAV surveying vs traditional methods. Im using a phantom 4 pro and Nikon D5300 with a 35mm prime lens

I took over 900 images on my 26 acre site. 300+ nadir images, 300+ in a oblique grid pattern, 150 in a oblique circular pattern and around 100 with a ground based dslr (at face level and elevated 3m on a painters pole).

My first chunk (nadir images) processed very nicely. I had 11 GCPs, I used 2 as check shots. Everything is in a local coordinates system (UTM10 but without the scale factor as my GCPs were shot in with a total station). 333/334 images aligned, control point error = 0.003m Check point error = 0.028m

I used this for my DEM and Orthophoto as it has been my experience I get a better ortho from nadir images only.

My second chunk (oblique grid, circular photos and terrestrial photos) initially wouldn't align well with reference mode due to the fact my circular photos had bad elevation metadata. So I aligned using the disabled option and 611/632 images aligned which I was happy with. Everything looked good when I turned the camera positions on. I built a dense cloud with mild filtering and was very happy with the results so I proceeded to input my GCPs. When I was finished and optimized my alignment I now had huge (35m) GCP error and now when I view my cameras, it is a total mess. I dont believe I did anything different and now cant figure out what my next step should be to find the issue.

I am using this methodology and settings (http://www.agisoft.com/index.php?id=31)

Any suggestions on where to start would be greatly appreciated.
« Last Edit: February 13, 2018, 12:21:57 AM by Shovelhead »

cadm8

  • Jr. Member
  • **
  • Posts: 74
    • View Profile
Re: Large Error in GCPs
« Reply #1 on: February 13, 2018, 03:46:06 PM »
Split your data in chunks by sensor type ie all p4p photos together and dslr ones apart. You can use different camera groups within the same chunk for the aerial photos, if you want to split them. Then realign chunks and follow the procedure for marking down GCPs.

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14816
    • View Profile
Re: Large Error in GCPs
« Reply #2 on: February 13, 2018, 03:51:14 PM »
Hello Shovelhead,

The first screenshot (oblique photos) demonstrates that the cameras are not correctly aligned. Not sure, though, if the initial alignment result or optimization, but anyway any further processing operations based on the incorrect alignment wouldn't give any valid result.

Also for oblique and terrestrial imagery I do not recommend to use Reference preselection option, use only Generic instead.

If you can provide the PSZ file with the alignment results of the oblique imagery (with placed markers) we can check it on our side.
Best regards,
Alexey Pasumansky,
Agisoft LLC

Shovelhead

  • Jr. Member
  • **
  • Posts: 75
    • View Profile
Re: Large Error in GCPs
« Reply #3 on: February 14, 2018, 09:25:27 AM »
Hello Shovelhead,

The first screenshot (oblique photos) demonstrates that the cameras are not correctly aligned. Not sure, though, if the initial alignment result or optimization, but anyway any further processing operations based on the incorrect alignment wouldn't give any valid result.

Also for oblique and terrestrial imagery I do not recommend to use Reference preselection option, use only Generic instead.

If you can provide the PSZ file with the alignment results of the oblique imagery (with placed markers) we can check it on our side.

Yes the first screenshot was trying to showcase the fact the alignment was off. I realigned my oblique set since then using the disabled option and actually had very good results. But the terrestrial photos did not align. So I am going try to try process those photos in a separate chunk and merge them in, however no GCPs are visible in the terrestrial photos. Hopefully it will work out. I am confused as why they aligned successfully before placing GCPs but not after when both times I used the disabled option of alignment.

Isnt the disabled option of alignment the best choice as far as accuracy goes vs generic or reference? Although it does take much longer.

stihl

  • Sr. Member
  • ****
  • Posts: 410
    • View Profile
Re: Large Error in GCPs
« Reply #4 on: February 14, 2018, 03:10:22 PM »
Isnt the disabled option of alignment the best choice as far as accuracy goes vs generic or reference? Although it does take much longer.
Generic down-samples the images before trying to find matches in the tie-points. This speeds up the alignment because Photoscan will know which images are taken consecutively to each other which speeds up the estimating camera locations step.
I'd only advice to use disabled mode either with small data sets (<100 photos) or data sets that will not properly align on Generic mode.
If you have geo-tagged images it's best to use reference mode. In this mode, Photoscan will use the coordinate information to predict which images are consecutive.

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14816
    • View Profile
Re: Large Error in GCPs
« Reply #5 on: February 14, 2018, 03:30:29 PM »
Hello Shovelhead,

The oblique images may not be aligned with the terrestrial photos, if the overlap between them is not sufficient to detect the common matching points, or if the view angle is too different and the objects, surfaces and etc. look very different when appear on the images from different sub-sets.
Best regards,
Alexey Pasumansky,
Agisoft LLC

Shovelhead

  • Jr. Member
  • **
  • Posts: 75
    • View Profile
Re: Large Error in GCPs
« Reply #6 on: February 14, 2018, 10:52:38 PM »
Hello Shovelhead,

The oblique images may not be aligned with the terrestrial photos, if the overlap between them is not sufficient to detect the common matching points, or if the view angle is too different and the objects, surfaces and etc. look very different when appear on the images from different sub-sets.

I get that, but why would they align before placing GCPs and not after? The images and overlap did not change at all.

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14816
    • View Profile
Re: Large Error in GCPs
« Reply #7 on: February 15, 2018, 01:34:35 PM »
Hello Shovelhead,

Markers' projections are acting as valid tie points  between the images they are placed on, so if the projections are not placed accurately, then it may cause the alignment issues.
Best regards,
Alexey Pasumansky,
Agisoft LLC