Forum

Author Topic: Seabed mapping -> alignment of 100.000 pics  (Read 1752 times)

tvi

  • Newbie
  • *
  • Posts: 8
    • View Profile
Seabed mapping -> alignment of 100.000 pics
« on: October 19, 2021, 01:37:44 PM »
Dear,

We are processing a 100.000 set of 1MB pics from an AUV.

As constraint we use rough position and more accurate motion data.

Alignment looks quite ok (all pics align), only issue we have is that the alignment consist out of different individual patches (See attachment).
What should be the best approach to align the patches to each other?

Regards,

Thomas

Bzuco

  • Full Member
  • ***
  • Posts: 144
    • View Profile
Re: Seabed mapping -> alignment of 100.000 pics
« Reply #1 on: October 20, 2021, 11:34:14 AM »
Do not know what is capable pro version of metashape, but using cloudcompare I would align two neighbour patches individually, then apply resample(it should suppress duplicates), then merge/align next patch and do resample, ...etc.

Or I would delete the common parts from one patch and keep that part only in second patch of the neighbours.
I do not know what exactly is on your screenshot and how it should look like without issue.

tvi

  • Newbie
  • *
  • Posts: 8
    • View Profile
Re: Seabed mapping -> alignment of 100.000 pics
« Reply #2 on: November 19, 2021, 01:44:16 PM »
Thanks for your answer Bzuco.

I would like Agisoft to do the alignment correctly. S
Something went wrong in the alignment stage.
An option to try and align/optimize different patches in one chunck was what I was looking for.

Currently I am now first configuring a MS Azure set-up for better processing times.
Then i can do a couple of reruns with different settings.

G,

Thomas

SimonBrown

  • Jr. Member
  • **
  • Posts: 59
    • View Profile
    • AccuPixel Ltd - Dealer and Training Centre
Re: Seabed mapping -> alignment of 100.000 pics
« Reply #3 on: November 19, 2021, 06:34:05 PM »
By "patches" do you mean separate chunks?
Agisoft endorsed online Metashape training - see: https://accupixel.co.uk/metashape-training-courses

tvi

  • Newbie
  • *
  • Posts: 8
    • View Profile
Re: Seabed mapping -> alignment of 100.000 pics
« Reply #4 on: November 19, 2021, 11:31:34 PM »
Hi,

No as you can see in the screenshot i have a single chunck with some "subchuncks".
All pics allign but these subchuncks do not.

G,

Thomas

CheeseAndJamSandwich

  • Full Member
  • ***
  • Posts: 151
    • View Profile
    • Sketchfab Models
Re: Seabed mapping -> alignment of 100.000 pics
« Reply #5 on: November 21, 2021, 10:44:50 AM »
Hi,

No as you can see in the screenshot i have a single chunck with some "subchuncks".
All pics allign but these subchuncks do not.

G,

Thomas
You know what it should look like...  I'm not sure what it should look like, or what your picture shows us...
Should it be just one long scan with a few parallel tracks???  The arrows are confusing...

tvi

  • Newbie
  • *
  • Posts: 8
    • View Profile
Re: Seabed mapping -> alignment of 100.000 pics
« Reply #6 on: November 21, 2021, 11:17:38 PM »
Ok,

Sry, i didnt suspect the screenshot came over as confusing.
It is a DEM of the sparse point cloud.

Basically I want 1 fully alligned chunck but the allignment is split up in several blocks.
You can see in the DEM that the blocks overlap but they do not connect (height jump).

I suspect that this is due to poor overlap while recording and or bad positioning at some part.
I also suspect some motion parameters which are not ok.

How can I most efficiently link these blocks together?
(With regards to the amount of pictures).
 
G,

Thomas

Paulo

  • Hero Member
  • *****
  • Posts: 1073
    • View Profile
Re: Seabed mapping -> alignment of 100.000 pics
« Reply #7 on: November 22, 2021, 09:06:18 AM »
Hi TVI,

it looks like there is poor overlap between adjacent blocks which would explain the height jump... Do you know what is the side overlap between adjacent runs?

Maybe select a photo from one run and use View Matches... from context menu to see if there are many or few matches with photos in adjacent run.... as in...

Another potential issue is very low resolution imagery (1 MB) which reduces possible matches... you can look at keypoint limit and also consider using Highest accuracy (Highest
accuracy setting upscales the image by factor of 4) for image alignment...
« Last Edit: November 22, 2021, 09:18:45 AM by Paulo »
Best Regards,

Paul Pelletier,
Surveyor

tvi

  • Newbie
  • *
  • Posts: 8
    • View Profile
Re: Seabed mapping -> alignment of 100.000 pics
« Reply #8 on: November 22, 2021, 09:52:38 AM »
Thanks Paul,

Switching on the thumbnail views makes everything unresponsive due to the amount of pics.

I will first try and upgrade the hardware.

G,

Thomas

Paulo

  • Hero Member
  • *****
  • Posts: 1073
    • View Profile
Re: Seabed mapping -> alignment of 100.000 pics
« Reply #9 on: November 22, 2021, 01:11:40 PM »
Hello tvi,

to rapidly check, you can share or look at 2 images, one from each adjacent parallel run to visually evaluate the overlap.... For reliable alignement you would need at least 50 to 60 % side lap
« Last Edit: November 22, 2021, 01:26:08 PM by Paulo »
Best Regards,

Paul Pelletier,
Surveyor