How to import object models using a scanner, and large buildings using Photogrammetry.
We recommend a modern PC (Ubuntu or WIn7) with 32GB RAM, newer CUDA Nvidia GPU cards (CC 5.1+, and 2GB+ memory), and the COLMAP Structure-from-Motion pipeline with built-in Multi-View Stereo reconstruction. This is one application that demands using the GPU.
While there is very limited OpenMVG/OpenMVS SfM support on the armhf OS for image processing, this benchmarks at over 80 minutes to reconstruct trivial 12-image examples on a pi3B+. Note, the Raspberry Pi will lock all cores at 100% use during processing, user interfaces will be mostly unusable during this task (systemd will complain in the error logs), and the OS can run out of resources if the image data-set is too large. Thus, final optional steps like "Dense Mesh Refine" surface improvements aren't currently possible as it would simply run out of memory on a Pi3B+ after several hours of processing.
Note, the Horus laser scan method will be faster on a armhf Pi, and in most cases should prove more usable performance wise.