Just finished scanning a packed pipe rack with photos only - here's what I learned

I ran into something similar when mapping out some large process vessels that had heavy corrosion. The sheer amount of surface variance made it difficult for the software to lock onto any reliable texture. It's not just the pipe geometry, either, but the surface quality, which really messes with the point cloud generation. For my job, I found that shooting in short bursts and then processing in batches was less painful than trying to get thousands of frames perfect in one session. If you have highly reflective metal, you need to be careful with direct sunlight because it throws out the matching entirely. I try to schedule the work for early morning when the light is soft, even if it means having to adjust the overall frame count slightly.

I found that instead of relying on surface features, using dedicated tie points or markers on the supporting steel structures worked better for ensuring the overall model scale was correct. It also helped maintain alignment when the pipe itself had inconsistent thickness due to wear. Since the work involves being on narrow walkways, I always make sure my tripod setup is stable and that all safety railings are clear before starting the capture.

When I ran a similar job at a fuel storage facility, the biggest headache wasn't the number of frames but the sheer complexity of the overlapping shadows. The lighting issue was definitly more intense than you suggest, because the metal and insulation cast huge, shifting dark areas that confused the registration of the adjacent pipe sections. I found that to get a clean result, I had to manually block out certain sections of the pipe before shooting them, focusing only on the exposed surfaces. The geometry was far more intricate than standard straight runs of steel; we had numerous spherical storage nodes clustered together, and the software kept misidentifying those spherical junctions as simple cylinders. I had to use very careful focal lengths, switching between a wide-angle and a 35mm lens just to capture the curvature accurately. The time spent adjusting those settings for every few meters of piping added up fast.

The repetitive geometry thing is brutal. I did a boiler house last year and the solver kept snapping sections onto the wrong parallel runs because every pipe looks like its neighbor. What helped me was walking a slow arc around each cluster instead of just orbiting flat, so there's some parallax for the software to chew on. Also masked out anything in the background between shots, sky, distant walls, all of it. Cut my alignment errors way down and processing went faster too. Did you find the upward shots through the tiers gave you enough parallax, or were they mostly just filling holes in the coverage?

Quick update — I did look into the shifting shadows you mentioned regarding the metal and insulation. If I were to process the raw data with alot of variance, did you find that using specific software settings for ambient light matching helped lock down the solver before the matching process even started?