I’ve tried searching for this with limited success… I assume its not possible!
I’m trying to generate a model using data from my Mavic 3M and my M30T, having a lot of success with the aerial imaging, but trying to combine with ground based images seems to be a problem.
I’m hoping to incorporate 360 images from an InstaX5 but the task fails every time when I’m trying to process those images onto heir own with the camera setting set to equirectangular. I don’t believe there is GPS tags on the images which could be the problem but what I’m looking to clarify is if it’s possible to upload both 360 and aerial images and process then at the same time for the same model.
Has anyone had any success with this and can share their workflow? I’m using both WebODM Lightning and Desktop.
Ground to Aerial “registration” is an active challenge that has no “clear” solution path; there are several streams of work that managed to certain success with different approaches. The main issue is to generate “features” (aka machine-generated “salient” points) that can be identified reliably across vastly different scales and viewpoints. That said, you could try using CGP (Control Ground Points) i.e. manually doing what the “computer can’t”.
In the interest of clarity – we are no longer the community of WebODM Lightning nor Desktop. There has been a split and that support is now at webodm.org.
-–
We do support OpenDroneMap Desktop and have general expertise on photogrammetry. I’ll attempt to address your question as it is a real challenge that is useful to have an open forum discussion on.
As to the technical problem of combining surface and ground imagery, Leonardo is right of course – this is an area of active research. Having successfully stitched across these boundaries, I would say a few things:
Unless Leonardo says otherwise, OpenSfM doesn’t easily support multiple camera types in one project. So the first problem you need to solve is ensuring that your 360 camera and drone imagery is in similar formats. In other words, the 360 imagery needs converted to (one option) projection to 6 cube faces, and then it can be combined with the UAV imagery, as a single brown (or automatic) camera type can then be used for both. Let me know if you need tips converting the panoramic imagery to cube faces.
Then there is the problem of stitching the full dataset together. We can think of this as an overlap problem similar to a UAV overlap/lawnmower flight planning process, but in the vertical. But it is probably easiest / best to just use common ground control point between the surface and UAV datasets.
Doing so solves the other problem, which is that OpenSfM is subject to model rotation when mixing camera angles under certain conditions, and this is one of those conditions. Ground control points will also solve for this problem.