Hello, we are new to WebODM and have run some tests on our images. we ran 5 images and it worked then 69 images also worked. We used ‘default’ settings (we think) however, there are some issues with the outputs so I have questions.
Running a new project what ‘Processing node’ do we use? Auto or node-odm-1? What’s the difference?
For ‘Resize Images’ do we do that? Yes or No? Would we not just keep our images original size for full resolution processing? I am unsure why we would resize?
The output geotiff image looks good but is offset about 10m from real-world location. We did this test (69 images) without any ground control so could that be the issue? Can we correct that after the fact?
I think I have asked too many questions for a first post! Sorry.
Our PC: Running Chrome, Windows 10 Pro, 32GB RAM, 100GB SSD free space. WebODM software was installed using the PC x64 installer without issues.
Hopefully I can help but take my answers with a grain of salt, I’m still an amateur.
Auto or node-odm-1 should give you the same result. I would assume that the auto is there for when you have multiple processing nodes so it would default to the node-odm1.
Resizing images helps with processing power. With less than 100 images you could probably leave it off. If you start to notice longer processing times then I would turn it back on. I haven’t noticed any large downgrades in quality from resizing.
The offset is almost certainly caused by a lack of GCPs. Depending on the type of drone you have there could be other issues. I have a Mavic Air and for a while I could not use it for mapping because GPS coordinates cut off too soon and therefore not accurate enough. With a new update to the API I could finally start mapping with it. As far as I know you can’t correct it after the fact but I haven’t actually tried myself so I can’t be 100% sure.
Thanks for your reply and thinking. Your comments sound logical.
I have re-run the same project (125 images) with the ‘resize’ option turned on - to see what, if any, difference there is in the output. The process took an extra 1.25 hours. I have not checked the output just yet.
Regarding the GCPs, yeah that makes sense as being the issue for the overall alignment. And we just do not have the ability to capture accurate ground control on site (we’re flying in PNG) so handheld GPS is all we have to go by. I am wondering if GCPs aren’t of any use to us in this instance. The project is to capture village layouts so cm accuracy is not really required. However, I was thinking I could simply create GCPs post-flying but your comment has me thinking otherwise. And, further, I was thinking to re-georeference the webodm geotiff again using other/base project imagery and it would suffice…Not sure though.
We have a Mavic DJI Pro2 (I believe) so we should be good to go with full coordinate readout. The test flights and stitched geotiff image look really good.
Hi,
I just wanted to give a comment about the GCPs: while it is best to do all the georeferencing in the ODM workflow, you can also use the georeferencer in QGIS with GCPs that you have created post-fly: for example, you can find roof edges and georeference according to them. You still need several points laid along the grid, but it should give you a good approximate precision on geolocation.
a (for example) 10cm offset is better than a 10m offset, and that may suffice for your goal
Thanks for those additional comments. I think georeferencing with GCPs created post-fly will be the best workflow since we do not need the absolute accuracy. Can you please clarify your comment “You still need several points laid along the grid…”. By ‘grid’ do you mean ensure we have a valid spread of GCPs (created manually) in QGIS using a coordinate grid overlay? My inclination is to create numerous GCPs (4-5) around the perimeter of the image set and a 2-3 down the centre.
Hi,
I meant the orthophoto itself. I assume you did not place targets on the ground during the mission, so you will need to find distinct features that can be accurately tag on the map. The Georeferencer tool in QGIS is very good at this
However, if you only need relative accuracy, your map is probably very accurate already. Try measuring distinct objects (like the length of a roof) to validate the map, but it should be pretty much accurate, especially if you aren’t looking for subcentimeter accuracy.
Point taken - I was thinking ‘grid’ as in an overlay grid or similar and not as a raster grid (I’ve always called image files rasters, knowing of course they are grids, but never refer to them directly as grids!). Anyway, I have georeferenced (using QGIS) the webodm orthophoto output and the image now aligns well. Probably within 0.2 to 0.3m accuracy - worse near the edges. Distances are still good, as you note, so that part is ok.
My desire is to re-process from scratch the source drone JPGs using a GCP file based on the newly-created GCPs. I read this post (Running OpenDroneMap · OpenDroneMap/ODM Wiki · GitHub) but I just can’t get my head around how to make the text file correctly.
For example, this image shows the my GCPs created in QGIS:
I am unclear how to (or if I can) get the file into the right format (and I would name it ‘gcp_list.txt’)?
I think I have to delete unwanted entries (i.e. enable,dX,dY,residual) and add the EPSG code at the first line. But how to add in each image number that has the GCP location visible in it?
