Odm for large scale aerial image orthorectification

Hey,
I am wondering if ODM is used or can be used to process (e.g. orthorectify) aerial imagery (shot with dedicated cameras from planes)?

In theory I imagine it to be quite similar, however I am not sure about this and would be interested if there are experiences. Possible differences to drone projects are e.g.

  • larger project areas (hundreds of km²)
  • lower spatial resolution (~10-40 cm)
  • larger images (tiles) >10.000×10.000 px
  • multiple bands at higher bitrange

Other software which is often used for this would be e.g. ERDAS imagine, Trimble inpho,…

3 Likes

Welcome!

We have had some folks try in the past with mixed results, but given our improvements around fast-orthophoto and planar SFM it might be possible.

Typically, such plane-borne sensor datasets have lower overlap/sidelap than our pipeline is tuned for, so it can be difficult.

Do you have any test data to try with, or to share so folks can take a poke?

1 Like

Thanks for your quick response.
Yes, the overlap will be lower. A typical overlap used to be around 60/30 when it was just about the orthoimage. However, due to increasing requirements for derived surface models and TrueOrthophotos (might be a german standard) newer flight campaigns often have higher overlap (e.g. 80/60). So we might approach from data and software sides.

I will have a look for datasets.

3 Likes

Hiya - I can test this next week with some old Hasselblad 50MP imagery I have around. I haven’t tried to push it through ODM for a few years now and was just the other day thinking to try it again.It’s not quite the scale of your imagery (‘only’ 8282 x 6240), it might give an OK indication :slight_smile:

4 Likes

This sounds much, much more promising!

1 Like

So I pushed a few (39) hasselblad 50MP images through ODM with … pretty decent results, considering the flights weren’t optimised for this (from the days where I had to push very hard to even run the camera so fast, because this SfM stuff was nonsense!)

Had no problem ingesting and working with the 50MP imagery. It is still a tiny dataset, seems more machinery will fix data bigness issues, everything in ODM just worked. Screenshots attached. This was run with a .geo file for camera centers, they’re pretty wrong, so DEMs etc are also way off.Will get another run done with GCPs in the next week or so… the GCPs I have are focussed on the buildings, so things will still get weird out at dataset edges.

I remember having more datasets - up tp 100 ish images, still not even scratching a real airborne campaign but maybe useful.

TL:DR, 50mp imagery works, don’t see a problem with pushing to larger images :slight_smile:

Also waaay back in the day we did our first proof of concepts on 20-40cm imagery (Nikon D1X flown at 500 m). I think as pixels get bigger, expect corners of things to be a little rounder.

4 Likes

Nice, thanks for your efforts testing this.

I am trying to push open source software here and was wondering if there is any suitable software to process aerial image campaigns. I dont know how this is done in other regions of the world but the situation here is that every federal agency has its own contract with different commercial providers (and as far as I can see no one is completely satisfied). So if it would be possible to achieve promising results with ODM or other os software it might be possible to use these and fund it for improvements.

It would be nice to have some kind of benchmark, do you think it would be possible for someone without ODM experience to get OKish results or is all about tuning the settings? We would have imagery to test this however i am not yet sure if I am able to share some of it.

3 Likes

I think it’s perfectly reasonable for someone completely new at photogrammetry to dive into ODM - especially webODM - and get OK / promising results :slight_smile: . Going from ‘promising’ to ‘ok we are now going to make legally binding decisions based on this’ is where the tuning comes in, same as for [insert commercial package here]

Of course I have a strongly vested interest in saying ‘yes hire experts!’ :smiley: - the reality is people will do OK pulling a few levers on their own most of the time.

…and at least my small exposure to the commercial / government landscape in Australia is similar - every agency gets contracted suppliers, and grumbles a bit. It’s also often a real issue to deal with the scale of data coming in, the impact of saying ‘ok we’re gonna hire someone to deliver 5cm imagery over 1000 sq km every quarter’. Thats another strong space for FOSS tools to operate in :tada:

3 Likes