Process huge mosaic, but process 3D mesh extracts later on?

Hi, I’m working for the local cadastre at the city of Rostock. Currently we evaluate, if we might offer more uptodate aerial imagery by using bigger BVLOS drone wings using cameras.
Due to the huge area (>350km²), I also see problems in stitching the photo mosaic.

While I guess we are able to process the photos and get an aligned DOP , I’m not sure about how to deal with a such a huge 3D mesh. Usually we don’t have an application that makes uses of 3D meshes.
Is there any way, how we can process the 3d mesh only for a dedicated area, for example to improve our 3D city modell for that region? This would save a lot of ressources, if it doesn’t has to be calculated for all the tiles…

Do you see any way, if we could work that way with ODM current? Or would it be a feature request with the focus to splitup the product generation for huge datasets?



Really neat use-case, and I love how you’re looking to have what I presume to be open-data driven by high-quality data, and open-source processing!

Currently, no, we don’t have a concept of a “Processing Area” (Pix4D parlance), but I do believe it has been requested before.

We also can’t do split/merge on the 3D model in our current pipeline, so it’d need to be processed at all once which would likely be a huge task with that extent of data.

The only thing I can think of would be subsetting your collected images to process just the needed area for 3D separately from the overall/complete dataset for the orthophoto.



I think you can skip the 3d model generation by using -skip-3dmodel.

-skip-3dmodel = Skip generation of a full 3D model. This can save time if you only need 2D results such as orthophotos and DEMs. Default: False

then you separately process the selected area data. It may will demand to adapt your existing GCP workflow. For example, once the orthophoto is processed you can select some features in the terrain to set some GCP and then reprocess only the desired area.

It’s not an automated process, but may work for your use case.



Thank you for the response, good to hear, that I didn’t missed an existing feature :wink:

Yeah getting the affected photos by a bbox selection would be the workaround, but of course wasting a bit ressources, because also this photos had been processed already for the full DOP. And yes georeference would be an issue, maybe a good point to think on RTK to avoid artifical GCPs…


This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.