Forum

Author Topic: Linear scan using video frames and RTK images instead of GCP's  (Read 4410 times)

wimdes

  • Newbie
  • *
  • Posts: 3
    • View Profile
Linear scan using video frames and RTK images instead of GCP's
« on: January 27, 2024, 01:56:00 AM »
Hi,

I'm trying to capture kilometers of road at close range to create a detailed 3D road surface model, using a Phantom 4 RTK.
With images this works very well (as they contain very accurate RTK GPS coordinates) but the drone can take only 1 pic per 2 seconds so the capturing process is way too slow.
Using video frames and sequential alignment works great but requires a lot of GCP's to keep the model in shape.
So I tried combining video frames with RTK tagged images every 5-10 or so meters as a substitute for GCP's.
This works *almost* but it's like the RTK images have less strenght in forcing the model into shape, compared to GCP's.
Is there a way to increase the 'weight' of RTK images in this workflow?
Any advice greatly appreciated!

regards,
Wim
« Last Edit: January 27, 2024, 02:25:21 AM by wimdes »

mrv2020

  • Jr. Member
  • **
  • Posts: 78
    • View Profile
Re: Linear scan using video frames and RTK images instead of GCP's
« Reply #1 on: January 28, 2024, 03:38:01 AM »
Hi,

Video usually has greater density with the extraction of frames, but it has a smaller format than a single photograph, you have to take this into consideration.

If you respect these parameters here,
https://agisoft.freshdesk.com/support/solutions/articles/31000149338-aerial-survey-tips,

Then you need to perform the alignment as here,
https://agisoft.freshdesk.com/support/solutions/articles/31000153696-aerial-data-processing-with-gcps-orthomosaic-dem-generation

If necessary, try to create random arbitrary manual control points to help the program identify the location correctly, selecting the images and alignment manually.
« Last Edit: January 28, 2024, 03:40:28 AM by mrv2020 »

JMR

  • Hero Member
  • *****
  • Posts: 533
    • View Profile
Re: Linear scan using video frames and RTK images instead of GCP's
« Reply #2 on: January 28, 2024, 07:40:57 PM »
Is there a way to increase the 'weight' of RTK images in this workflow?
Any advice greatly appreciated!

regards,
Wim
Hi Wim:
Maybe is worth to mention that the weight of camera locations can be set according to the quality of the RTK fix solution.
If your drone with RTK on board can write this tags to the images, then you must tell Metashape to read them from metadata or alternatively you can enter all manually by selecting all cameras on ref pane, right click and choose modify accuracy, then enter the estimated accuracy.
a single value is assigned to xyz
two values sepparated by / are assigned to xy/z
three values are assigned independently to all axis x/y/z
To tell Metashape to gather the values from metadata, goto tools/preferences the got advanced tab and make sure that "Load camera location accuracy from XMP meta data" is enabled.
If it was not enabled for your project, just re-read metadata by means of the button with a camera, (third from the left) to "Import Reference from EXIF"

José Martínez
Accupixel
https://accupixel.co.uk/courses/metashape-professional-edition-for-forensics-surveying/

wimdes

  • Newbie
  • *
  • Posts: 3
    • View Profile
Re: Linear scan using video frames and RTK images instead of GCP's
« Reply #3 on: January 29, 2024, 02:02:09 AM »
Exif data is imported and accuracy values do show up in image reference.
Alignment works by
- selecting 'sequential preselection' for the frames and including the first (and optionally the last) RTK frame in a first alignment.
- then adding the other RTK images using 'generic preselection'
All imges align properly, the model will at first be deformed, by 'optimizing camera's' a few times the resulting model gets close but not 100% where it should be

wimdes

  • Newbie
  • *
  • Posts: 3
    • View Profile
Re: Linear scan using video frames and RTK images instead of GCP's
« Reply #4 on: January 30, 2024, 09:38:39 PM »
RTK accuracy is 1-2cm.
Using about 40 RTK tagged images on a test stretch of 250m, after optimizing camera alignment a couple of times, reported error on the pics ranges from 6 to 88cm.