Comparing Orthophotos

I update progress on a nearby construction site, including mapping, and have been producing orthophotos using Drone Deploy. I initially compared those results to WebODM orthophoto with all-default settings and was disappointed that DD was clearly a better image. Then went back into WebODM and re-ran the same file with the Hi-Res template and found that it was maybe a little better than the DD product. Nice!

After reading the Blog recommending the Brown-Conrady lens option as an improvement over the default lens, I again reran the data set with the Hi-Res template edited to the Brown lens. I found the Brown lens to be better in some ways than the default and not quite as good in others. Can’t characterize a compare/contrast, but they were just slightly different to one another and both seemed slightly better than the DD product in many ways, although DD had a very nice orthophoto.

If anyone is interested, I’ve uploaded all four orthophotos to dropbox at this link: Dropbox - File Deleted - Simplify your life

I’ll also upload the data set to the ODM data pool if anyone wants to work on it themselves. This construction site offers a rich set of features for comparison. Lots of different textures, some vegetation, some reflections on water, lots of different colored items that should be straight lines, some highway, cars, structures, etc. This project is known as Miracle Park and I’ll label the file that way. It’s 122 JPGs, and ran in WebODM on my dual-quad core 60GB PC with these times (H:MM):
All defaults : 1:40
Hi-Res template: 3:43
Hi-Res with Brown lens: 3:35

The orthophotos had these GeoTIFF file sizes:
DroneDeploy: 678MB
WebODM All-Default: 154MB
WebODM Hi-Res Template: 867MB
WebODM Hi-Res with Brown: 867MB

Offered in case anyone else is searching for answers to orthophoto resolution questions. Images recorded with Phantom 4 Pro. Best regards… Bob R.

3 Likes

A couple of suggestions for starting a comparison:

  • check the differences in the reflections on the standing water in the detention pond
  • check the ladder on the side on the truck in the different images
  • check out the linearity of the blue plastic pipes laying on the ground and the metal roof on the fire house in the upper left corner.
  • the track hoe is kind of interesting also. The tracks and arm didn’t resolve well in any of them.

Sorry I don’t have the chops to put these into the Arena-style slider. You’ll just have to open the images and click back and forth.

2 Likes

Great info. Thanks, Bob!

Bob,

I played with the dataset and WebODM parameters to produce a 3D Model and Orthophoto.
Here are the parameters used: mve-confidence: 0.4, texturing-nadir-weight: 32, orthophoto-resolution: 1, dem-resolution: 4, mesh-size: 300000, mesh-octree-depth: 11, dsm: true, camera-lens: brown, depthmap-resolution: 1000, texturing-data-term: area.
This is my first time using the Brown lens option.

I have been testing odm parameters to have high resolution models and orthophotos just like those resulting with other software. I must admit that creating a nice orthophoto in urban environment using WebODM is somehow challenging.
I expect some critique on these parameters.

Processing time for these settings was 01:05:05 using a WebODM Lightning Processing Node.

The resulting 3D model is good, at least for observing the excavation works.

I compared results from DD and WebODM, on these links you will find the comparison, orthophoto and the textured model.

Comparison:

Orthophoto:

Textured model:

Thank you for sharing

3 Likes

Hello Israelbar – found your post just as I was winding down for the evening. Will enjoy looking through your work this weekend. Hope you found the data set useful. Have you drawn any conclusions from your study?

Bob R.

1 Like

Thank you for the great info @HAWK-i.us.

@israelbar, what are the specs on your computer? The runtime is great.

It is my understanding that the DEM is used to create the orthophoto. If so, does the miss-matched resolution between the DEM and orthophoto resolution make a difference? orthophoto-resolution: 1, dem-resolution: 4,

Thanks,
CJ

Hi,

It was processed using WebODM Lightning. I strongly recomend you to try it.

And regarding the miss-matched resolution between the DEM and orthophoto, I think DEM resolution has no impact on orthophoto resolution since it is generated by taking a picture of the textured 2.5D or 3D model. I’m just understanding how ODM works so maybe I’m wrong.

Saludos

2 Likes

Spot on. The difference in resolution is a function of the practical limits of photogrammetry: a basic estimate is if your sampling resolution if 1 (typically approximately the same as your orthophoto resolution), your elevation resolution is a good balance of noise and information at 4. This isn’t universal, but it’s a good starting place for results.

2 Likes

I’m processing a task on Lightning Network as I type this.

This from “The Missing Guide” confirms your statement.

Orthophoto Processing

The orthophoto is generated by taking a picture of the textured 3D mesh from the top.

WikiPedia

The Orthophoto WikiPedia Orthophoto - Wikipedia is where I saw the dem is required for an accurate othophoto…

A digital elevation model (DEM) is required to create an accurate orthophoto as distortions in the image due to the varying distance between the camera/sensor and different points on the ground need to be corrected.

I’m just trying to figure this out.

cj

2 Likes

uametry – I share your pain with the learning. It’s a “hill for a stepper”. The “Missing Guide” is an excellent resource. The Community thread here has more background: Where can I find background information on the concepts of ODM? - #2 by pierotofy

2 Likes

Seems that smathermather-cm has answered our questions.

How was your experience using the lightning Network?

Saludos

It was delightful @israelbar.

I tried processing a set of 350 images on my macbook pro with 32GB of ram setting the ortho_resolution to 1 and I let it run for somewhere around 60hrs, I killed the process.

I uploaded 433 images to the lightning network and set the ortho_resolution to 1 and the processing took 06:35:32. The quality of the outputs are very good.

CJ

3 Likes

Got similar results here, I processed 980 images in less than 8 hours.

For most of the projects WebODM + ligthning is the way to go.

Saludos

3 Likes

:sunglasses: :+1:

If the requested orthophoto resolution is a higher resolution than the Ground Sampling Distance, does WebODM just limit the resolution to GSD without notifying you, or what does it do (assuming that the Ignore-GSD option is not set to TRUE) ?

Where do we find what the computed avg GSD is in the output?

3 Likes