Map wizard fails to align forest images. Looking for advise on tweaking settings to reach 100% alignment.

Development Asset Creation
1

Dear support team, a newbie question, but still…

I’m running a forested area mapping project with aerial photos (12mpix, 70% front and 80% side overlap). There are hundreds of images, out of which over 50% won’t align with auto settings. The ones aligned contain roads, buildings etc. but the ones with only vegetation - fail.

I understand that forest is a difficult area, but other software somehow manage to get it done with more “relaxed” settings.

Please help with “forest” settings so I can finish it with RC. Thanks.

2

Hi chemist207,

forest is always tricky as there are quite big changes between images, so it is hard to find common features between images.

There could help also capturing from higher flight height. What was your flight height?

Also, you can try to set bigger values for Max features per mpx, Max features per image and Preselector features in Alignment settings. You can also set higher number in Max feature reprojection error and set Low Image overlap.

Do your drone images have position information? If so, you can set lower values in Position accuracy (maybe to 1 m) and also higher value for Position prior hardness. Also, you can set higher Detector sensitivity.

Is it possible, that you used RTK drone? Then you can also use flight log.

3

Thnks, Ondrej. Will try these settings tonight.

The altitude was 300m. Unfortunately, RTK kit is currently out of scope for me.

The drone is Autel EVO II, shots 12MPix

4

Hi . Tried every suggestion from your post - the max aligned points is 150 out of 800.

Looks like i’m missing something critical. Could you please check my settings or maybe poin out to some comprehensive reading about this topic?

5

Hi, is it possible to send us your data for some tests?

But as I wrote, forest is quite hard dataset to align. What was your overlap between images?

6

Hi there. Thanks for qick reply. The overlap is 70% front and 80% side.

The dataset is quite heavy. I can send a link to a fileshare to your email. Is it OK? If so, please share your email via PM maybe.

 

How about the settings I used? Anything to be tweaked further? I managed to get this amount of points aligned:

7

Another thing is that from the aligned cameras many points were generated outside of actual srface areas. Not sure how to deal with this.

8

Hi,

the setting above are not ideal. There is quite big value for Max features per image. I don’t recommend to use values higher than 80 000. Also, max feature reprojection error is enormous, maybe 5 - 6 could be useful there. Also I would consider higher value of Position prior hardness (maybe 10)

I will send you a PM from mail mail address and then you can send me a link to download the dataset.

Regarding to the showed issue, this is a result from wrong alignment. You can choose these wrong points and disable cameras, which are related to them. But it is not ideal More I can say after I will see the dataset.

9

As I see your aligned images, maybe also double grid could help here.

10

Double grid shold definitely help, but it will not only double the abount of data to process, but also is hard to achieve in one day because of flight time and number of spare batteries on hand :frowning:

11

Yes, that is true. I was testing that dataset and wasn’t able to align it correctly. Also, it could help using some GCPs placed on the ground to align these images.

This could be also helpful, but it is an article for different issue: https://support.capturingreality.com/hc/en-us/articles/360018811880-Banana-effect-what-to-do-if-my-model-is-bent