Warping of trees and buildings

optimization

#1

I am doing orthophoto work - vegetation and geology work - and have managed to get my images pretty good now. I am still having a few issues with warping of trees and buildings. I changed the mesh octree depth to 6 then 3, then 1 and finally got images that were a lot less warped around the trees. any suggestions to help would be greatly appreciated.

Other settings:
Fast orthophoto on
min num features 50000

I have included two snapshots from on of my images to show what I mean.

Thanks for any help you can provide :slight_smile:
odm_orthophoto_cropped_02


#2

Hi All,

So I have been looking at why my data was doing what it did. It took me a while, but I realised that the data had big gaps. It was in the gaps that the warping was occurring.

I looked at my flight lines and realised that my drone had skipped lines in some areas (about to mail developer about this). The warping was resulting from a less than 60% overlap.

In my most recent processing I managed to get only a handful of trees warping by adding in some infill photographs from a previous flight. My minum features were set to 20000 fast orthophoto was on and mesh-octree was 6, so it clearly has less to do with the settings than the photo density.

I will re-run my flight when I get a chance and increase the overlap of images so that I will have a 60% overlap if my drone misses lines and I will post back with results. I hope this information will help people that have the same kind of issue that I have had.


#3

Excellent, @Pat. Keep us updated. There’s also some work underway the next few weeks to improve orthophotos, so it will be interesting to see how much that helps as well.


#4

Hi All,
I faced similar problem. I’m tried different processing parametrs, but issue still persist.
Here is my best results:
Settings:
mesh-octree-depth: 12 (also tried with ‘6’)
mesh-solver-divide: 12 (also tried with ‘6’)
resize-to: -1
orthophoto-resolution: 100
force-ccd: 13.1
min-num-features: 50000

How can I improve resulted quality? Thanks in advance:slightly_smiling_face:
test


#5

Hi Artoum,

My main issue turned out to be my photo overlap. for some reason that my drone manufacturer still doesn’t understand my drone was missing lines in the survey. To solve this I simply need to do an 80% side overlap. This will guarantee me a 60% overlap if my drone misses a flight line. I added extra data from a first test flight and it improved my imagery to the point of being useful for doing my vegetation work.

There was one extra drop down menu that I changed near the bottom of the settings menu called dem-terrain-type. I changed it to flat forest for my work, but you could use complex non-forest for your processing and see if that helps.

The settings I changed on the final image I produced were:
mesh-octree-depth: 1,
min-num-features: 25000,
fast-orthophoto: true

knowing about the overlap issue I would run my survey again and ensure better than the minimum required overlap. I would then re-run the orthophoto only changing the mesh-octree-depth to 6. I just haven’t had a chance to re-run my survey yet.

Your image does look a lot like there is a lack of vertical images for the correct unwarping and interpolation. What was your image density like?

Pat.


#7

Hi @Artoum_Gulyaev – as alluded to earlier, we do have some fixes in the pipeline to substantially improve outputs, which address precisely the issues you see here. See https://github.com/OpenDroneMap/OpenDroneMap/projects/2 for some details on the pieces needed to improve.

Our expectation is to release these improvements 2 July or so, but there are some logistics to that release that are not under our control, so we don’t yet know the exact timing.


#8

Hi @Pat,

Many thanks for your answer and yes, I think, you were right. I processed my set of images with Pix4d and faced similar problem. Now I will try to capture more images and retry my tests. Btw, does ‘fast-orthophoto’ option affect the final quality?

Hi @smathermather-cm,
Thank you for informing me. I’ll look forward to the next update.


#9

Rarely does fast-orthophoto increase quality. This is an approach that is an attempt to shortcut some of the photogrammetry steps to speed things up for users who need speed over quality. There are some use cases where it gives very good results, but should not be relied upon for this.


#10

@smathermather-cm Any updates on it? Also, when do you plan to release a new version?


#11

Hi All,

It has been a little while since any update has been posted on here so I thought I would mention some things that have helped me:

  • The new update run on old surveys drastically improved the output. Thank you for the updates. I really appreciate the time and effort people put in.
  • I have a cousin that works in photogrammetry and he suggested flying two altitudes to my surveys. This was a big leap forward for me. All tree warping disappeared in the area of interest in my surveys with this additional step. To do this if you fly at say 30m you would then fly the second one at say 45 or 50m (the surveys need to be close enough together so the software can match the features on the images still). This is to give better continuity over images and gives a second set of better than 90% overlap. I did three surveys and used 45, 60 and 90m elevations.
  • Another survey method I have been reading about recently has been the flying a second survey with the camera not viewing nadir. @smathermather-cm mentions this in other posts. This adds more depth to the point cloud is my understanding (correct me if I am wrong), improving the final output.
  • I am currently re-running my dual survey because I still have issues with the buildings even with the two sets of images and continuous straight edges. I am running with use fixed camera parameters turned on and if this doesn’t improve the building edges I will unwarp the images prior to processing.

I hope that people can use the information that I have provided from my experiences to help improve their surveys and output products.


#12

Thanks for sharing @Pat! I think we should start considering opening a wiki of sort to collect all flying tips into a single location.


#13

I think that would be a good idea. It would help a lot for the beginners like me out there that are trying to produce a quality product for their own personal interest and development. There are a lot of small changes that help the survey results a lot and many of them aren’t documented in one place.


#14

I wouldn’t use fixed camera parameters unless you have unwarped the images first. Otherwise, this is a really useful summary. Thanks!


#15

The fixed camera parameters didn’t improve results. I will look at unwarping my images to see if that helps with the distorted structures.


#16

Yes, fixed camera parameters won’t help without undistort. :crossed_fingers: Fingers crosses for your next results.


#17

How do you plan to undistort your images?


#18

I was going to use lensfun in hugin on windows. wanted to use it in gimp but there are issues there that I cannot fix.


#19

Are they JPEGs or RAW?


#20

jpeg files.


#21

outliers%201 outliers%202 outliers%203 airphoto%201

these are the sort of results I am getting.