Incomplete Ortho's

[JedW]

I’m consistently getting very incomplete Othomaps, particularly in wooded areas and am at a loss as to how to fix this:

SETUP:

• Autel Evo 2 Pro flown at 200ft, lawnmower pattern, 85% front overlap, 75% side overlap, camera pointed straight down (90degrees)
• Processed using WEBODM - Lightning network
• First attempt used Fast Ortho… all default settings
• Second attempt used High Resolution… all default settings

Results
Both outputs cropped about 25% of the woods to the east out of the othomosiac.

Here is a screenshot of the autel mission map:

And here is a screenshot of the output othomap:

And here is the Processing Report:
report3sm.pdf (1.6 MB)

This is a pretty consistently occurring problem for me so any assistance would be GREATLY appreciated.

Thank you.

[Andreas_Olsson]

Was it windy when you did the flight? Moving leaves isn’t good at al.

Try not to use 3D model for ortophoto and cropping the images when processing I think might help.

[JedW]

The wind was 5-10mph… I think that’d be about 8-16kph.

First time I tried processing it was as a “Fast Orthophoto” which I thought bypasses the 3D model… it produced the exact same output

Sorry for the noob question but how do I try cropping the images when processing?

[Andreas_Olsson]

There’s a setting for detail, default is “normal”. I think it crops the images.

[smathermather]

@JedW

85% front overlap, 75% side overlap

I find 83% (front and side) the magical number for forests, it’s what is necessary for true 3D reconstruction which makes it easier to map low height flights over complicated canopy tops.

That said, you can turn up the knobs on matcher-neighbors (make it 20 or more) and turn up the knobs on min-num-features to something like 32000 or 64000. You can also try setting feature-type to orb. I haven’t had luck with this yet, but I know others have, and if it works it will speed up your processing.

(I don’t know if there’s a cap on lightning for min-num-features, but if there is, set to to the cap instead of the above)

[Andreas_Olsson]

Doesn’t overlap depend on the hight to width ratio of the sensor?

[smathermather]

Yes, and most flight planners take this into account.

[Andreas_Olsson]

Haven’t seen that in Pix4D Capture.

[smathermather]

Interesting. I am pretty sure the FOSS ones do. I am surprised that Pix doesn’t, although it simplifies things quite a bit to ignore this.

[Andreas_Olsson]

How do you know?

[smathermather]

How do I know what?

[Andreas_Olsson]

How do you know if the flight planer takes image ratio into account?

[smathermather]

I am fairly certain that when I was regularly using Mission Planner, you could specify whether the camera’s longer axis was rotated inline or at 90 degrees from the flight line, and the flight plan and spacing would change based on that parameter. Mission Planner gave you a full IFOV calculation with number of overlap estimates as well. The IFOV calculations definitely took into account height and width ratios.

Anyway, we’ve hijacked this thread… . Sorry @JedW. Check out the advice above for either tweaking parameters or reflight overlap recommendation.

@smathermather

I find 83% (front and side) the magical number for forests, it’s what is necessary for true 3D reconstruction which makes it easier to map low height flights over complicated canopy tops.

That said, you can turn up the knobs on matcher-neighbors (make it 20 or more) and turn up the knobs on min-num-features to something like 32000 or 64000. You can also try setting feature-type to orb. I haven’t had luck with this yet, but I know others have, and if it works it will speed up your processing.

(I don’t know if there’s a cap on lightning for min-num-features, but if there is, set to to the cap instead of the above)

[Saijin_Naib]

In addition to Stephen’s advice, you should pass --crop 0.

For ORB on Lightning, I’ve gone up to 1.2M features, but it was worse than 120k features for ORB.

Also, in future flights see about setting 75-85deg for the gimbal.

[Gordon]

You could try the forest setting, which bumps up min num features to 18000 I think. For me that has produced satisfactory results in densely treed areas, but for less dense cover, 10000 has worked.