Photogrammetry
Large-scale terrain scans via aerial drone.
About
Several years ago, I took a summer science course at UCLA, where we used photogrammetry scans to study local plants.
Later in my last year of college-- when I learned that professors at my university were interested in building a 3D asset library for student use-- I took this experience and applied it to game asset development. I proposed an independent grant in which I'd create large (<20 sq. ft) photogrammetry scans using a drone.
In January 2026, I received funding from my university's project committee for my project. The project took four months to complete, and I presented my results at the university's end-of-year awards ceremony.
Scanning
I scanned in the wider Massachusetts area; specifically, in parks and nature. I waited for clear, bright days and performed scans at noon when the sun was at its zenith, for top-down lighting.
Choosing subjects was its own learning curve; I made several mistakes during the scanning stage. I quickly realized that taking pictures near water (or any reflective surface) would create patchy scans. Likewise, any terrain that was too grassy or prone to moving in the wind made inconsistent scans.
I avoided these problematic subjects and had great success scanning rocks, cliffs and gravelly terrain.
Point Cloud Rendering
After gathering my images via drone photography, I compiled my assets in RealityScan (Epic Games). The software gathered anchor points to generate a mesh from point clouds, which looked excellent at a glance, but posed several glaring issues.
Firstly, the mesh was inconsistent (variable density of polygons) which slowed down performance. Secondly, the mesh was full of holes, which would cause issues for any artist looking to work with it.
Since I was creating these scans for other students, I didn't want to provide them with a difficult or messy asset. I needed to clean up the scans.
Cleanup
I used Houdini to clean the raw meshes into workable assets. I deleted all all edges above a certain size to effectively filter by detail (larger polygons are less detailed in scans), then isolated non-manifold topology and filled the gaps. Finally, I remeshed the entire surface, re-baked the texture maps, and made the UVs 0-1 tiles.
Conclusion
In the end, I produced three high-quality large terrain scans for student use. They're stored in a shared library that's accessible to all students of my college. I enjoyed the project, especially the procedural cleanup process (working outside was fun, but keeping the drone safe was stressful! I prefer sitting at a warm, rain-free indoor desk). I was able to keep the research drone, and intend to work more with photogrammetry in the future.