OpenDroneMap / ODM

A command line toolkit to generate maps, point clouds, 3D models and DEMs from drone, balloon or kite images. 📷
https://opendronemap.org
GNU Affero General Public License v3.0
4.71k stars 1.08k forks source link

Orthophotos mismatch at border areas #652

Closed hy9be closed 6 years ago

hy9be commented 6 years ago

Hi! I am trying to follow the "splits my dataset into 80 manageable chunks" strategy discussed in smathermather's article.

image

I tried to group images into different tasks under the same project in WebODM. But the result I got has some mismatch at the borders of orthophotos from different chunks, even they actually use the same image files at those area (see the image below).

Any suggestions to improve? Thanks!

image

KommandorKeen commented 6 years ago

Did you have overlap between the groups? See attached sketch Each group needs to be trimmed at constrained rather than unconstrained borders.
file 17-8-17 18 57 20

hy9be commented 6 years ago

@KommandorKeen I do not have that many overlaps (only 5 instead of 15, as below). And all my images are taken vertically.

image

KommandorKeen commented 6 years ago

OK, try overlapping the photos so that g1 and g2 share 2 rows of photos then trim orthomosaics so they join between the two shared lines. If that is more promising, but not perfect, make g1 and g2 share 3 rows of photos then trim the orthomosaics along the line of the middle overlapping rows.

hy9be commented 6 years ago

Basically try to have more overlapping? Am I getting it right?

I do not quite understand how the two tile can mismatch at the border, as they share the same image files. Could you give me some ideas? Thx!

KommandorKeen commented 6 years ago

Think of all the images stitched together as a big rubber sheet. The sheet is stretched out by all triangular relationships between cameras and identifiable points on the images. Nothing actually fits perfectly and the tension is uneven across the sheet. Each camera position is well defined by links to many other cameras and image points, Once you move outside of the area where tension on a point can come from opposite directions (looking down - has a camera on 2 of more opposing sides of it) then the position is less constrained and the rubber sheet can warp in unexpected ways. Each group of images have different tie point relations and therefore different tensions causing different distortions. Hope this helps.

hy9be commented 6 years ago

Thanks @KommandorKeen !

hy9be commented 6 years ago

@KommandorKeen I read several papers today and I think I got the idea. One more question (maybe a pretty dummy one..), by saying "trim orthomosaics so they join between the two shared lines", do you mean I should manually edit the orthomosaics?

KommandorKeen commented 6 years ago

Yup, if you load the orthomosaics into qgis and use the geotag and import photos plugin you can see where the photos are and create a polygon (or series of polygons) to trim to. using the raster - extraction - clipper function

smathermather commented 6 years ago

@hy9be -- you might try these scripts to split and re-merge your dataset: https://github.com/paulinus/OpenDroneMap/tree/feature-split-dataset/scripts/metadataset. They're not in master yet, but they work quite well. See also: https://www.dropbox.com/s/7c3b3k4us2t8e3i/demo_zanzibar.mp4?dl=0

hy9be commented 6 years ago

@smathermather I tried the ./run_all.sh but runs into some runtime errors. Should these scripts totally replace the reconstruction steps in odm_app?