Photogrammetry with Google Earth Studio
Recently explored the potential of Google Earth Studio as a tool for planning and previewing drone photogrammetry workflows. Around the same time, I was interested in creating small 3D-printed models of notable infrastructure at NASA Ames, such as the NFAC (the world’s largest wind tunnel). Since Google Earth already provides detailed 3D terrain and structure data, I thought it would be interesting to test whether Google Earth Studio could serve as a simulation tool for photogrammetry.
Getting started was fairly straightforward. After generating a short animation, I used RealityCapture to convert the footage into a textured mesh, which I then cleaned up in Blender and prepared for 3D printing. As a secondary experiment, I also used Jawset PostShot to generate a Gaussian Splat from the same footage. While the splat wasn’t perfect due to limited sky coverage in the animation, the overall results were compelling. Here’s a quick overview of the workflow and results.
Step 1: Google Earth Studio – Generate Animation
Google Earth Studio is a browser-based animation tool from Google that allows you to create cinematic flyovers of real-world locations. Documentation is available here. To access it:
- Go to the Google Earth Studio website
- Click Try Earth Studio
- Sign in with a Google account and briefly describe your intended use
I used the spiral orbit template centered on the NFAC. It’s possible to import .kml files into Earth Studio—though based on my testing, these files only support overlays and paths for visualization within the animation, rather than actual flight control.
Google Earth Studio templates
Google Earth Studio main interface
I exported the rendered animation as an .mp4 file, which could be directly imported into both RealityCapture and PostShot.
Animation flyover of the NFAC at NASA Ames. Attribution: Google Earth, Vexcel Imaging US, Inc.
Step 2: RealityCapture – Convert Video to Mesh
In RealityCapture, I imported the .mp4 and allowed it to extract approximately 200 frames. Using mostly default settings, I ran the photogrammetry pipeline to reconstruct a textured mesh.
Generated textured mesh from Google Earth Studio footage
After reconstruction and simplification, I exported the mesh as a .obj file for further editing in Blender.
Step 3: Blender – Mesh Editing
Photogrammetry-generated meshes often require cleanup. I used Blender to refine the geometry and make adjustments to prepare the model for 3D printing.
Post-processing and cleanup of the mesh in Blender
Step 4: Orca Slicer – 3D Printing
The refined mesh was then imported into Orca Slicer, where I set the model size and printing settings.
Slicing the model in Orca Slicer
Step 5 – Print Results
The final 3D print is shown below. The full model is roughly 12 cm across, with the wind tunnel itself clearly recognizable. Some areas could benefit from additional mesh refinement to ensure consistent flatness. Surrounding facilities could be better defined but are in most part still identifiable. One idea for future iterations would be multi-material printing to better distinguish between buildings, roads, and grass areas.
Final result: 3D printed model of the NFAC
Extra: Jawset PostShot – Gaussian Splatting
For comparison, I also processed the video using Jawset PostShot to create a Gaussian Splat. I used the splat3 radiance field profile and left most settings at default. The tool extracted around 150 images and reconstructed a lightweight splat-based rendering.
One benefit of splatting is that it can represent background elements like the sky—something that traditional photogrammetry struggles with due to a lack of trackable features. However, because the animation included limited sky coverage, some regions remained unrepresented.
Gaussian Splat generated from the same animation using PostShot
Final Thoughts
Google Earth Studio is a surprisingly capable tool for generating flyover-style animations, and it integrates well with both RealityCapture and Blender for experimentation and prototyping. I was able to produce a reasonably accurate physical model of the NFAC, entirely from virtual footage.
As a future improvement, it would be useful if Earth Studio supported importing .kml flight plans for true photogrammetry simulation. Although .kml overlays are supported, they currently don’t drive the camera directly.
Interestingly, the watermark embedded in the rendered video didn’t show up in either the mesh or splat results—likely rejected as noise by both systems.
For those without access to drones or physical sites, I recommend exploring freely available photogrammetry datasets such as those hosted by ESRI. They’re a great resource for tuning your reconstruction pipeline before working with your own data.
Tools Used:
Enjoy Reading This Article?
Here are some more articles you might like to read next: