Best Workflow in Meshroom for Drone Images with Geolocation (Solar Simulation Use Case)

[GabrielSk82020]

Hello,

I’m working on generating 3D models from drone images using Meshroom, and I’d like some guidance on the best workflow or pipeline configuration. Our main use case is for solar energy projects, where we simulate shadows cast by surrounding objects and evaluate optimal placement for solar panels on buildings.

Since geolocation accuracy is critical for our simulations, I’d like to know:

• What is the recommended pipeline in Meshroom when processing drone imagery with GPS data?

• Are there specific node settings or best practices to ensure both high-quality 3D reconstruction and precise georeferencing?

• Any tips to balance quality with performance (so we don’t run into excessive RAM usage) would also be appreciated.

• Is ou

Additional Context

• Drone used: DJI Mini 3

• Pipeline: Default nodes with some tweaks to reduce 3D model size

• Issue observed: In one of my current projects, the shadow in the 3D model appears shorter than the real-world shadow. This makes me suspect that the GPS coordinates from the drone images might not be fully accurate, or perhaps not being used by Meshroom at all. For this test, I exported the model and visualized it in SketchUp Pro 2019.

Thanks in advance for any insights!

[GabrielSk82020]

I discovered the issue was actually related to the coordinates. In SketchUp, there’s an option to manually insert them, and after doing that, the shadows aligned correctly with the real ones for that specific date.

I also found out that Meshroom doesn’t use GPS data from pictures by default. However, there’s a node that should fix this: SfMTransform → from_gps.
I tried using this option, but something went really wrong. The 3D model didn’t show up in Meshroom’s 3D Viewer. Suspecting it still generated something, I exported it to .obj. In Blender, by using the numeric keypad dot, I located the object, but it was very far from the origin point (0,0,0). Unfortunately, it looked incorrectly processed — completely deformed, almost like a flattened pancake.

Has anyone successfully used this node? If someone could share a working .mg file with the correct pipeline, I’d be really grateful!

[LordMaddhi]

Maybe this helps: #2691

[natowi]

The 2025.1 release will have a new functionality for "Survey point injection"
Inject survey points (3D world coordinates + 2D coordinates in a given frame) using a json file generated from 3D equalizer. Add survey points to sfmData and use survey points to constraint the pose estimation in the bundle adjustement.

But to be honest, the most fitting software for your use case would be Open Drone Map (ODM)