Well, do you know where these shelves are? I'd be interested in photogrammetry software to reconstruct 3D models of environments, mostly to plan construction or aesthetic changes to buildings or surroundings. This is for home use, so I don't really need fancy texturing, etc. But having it reconstructed from a smartphone video feed would be a must.
I looked for open source software a while back, but didn't really turn out interesting results :/
Edit: there seems to be a few interesting resources. It looks like MicMac could be quite simple to use, and desn't have a CUDA dependency, for instance.
It's been a few years but in my photogrammetry days the open source options were very unfriendly to use -- either a command line utility lacking documentation or an extension for an outdated version of something else, and I never got a result I was happy with. it can take a couple of hours to process large photo sets and it's really frustrating to get a bad result.
For $180 (non commercial license) Agisoft Photoscan worked the first time and gave me lots of tools to get a good, meshed and textured result that I could export to other software for viewing.
Smart phone cameras are actually ideal because their tiny sensors give huge depth of field -- everything is in focus in every picture == happy stitching
EDIT: you'll be waiting a long time for alignment without GPU !
If you want to investigate free software options then may I suggest you look into Visual SFM (http://ccwu.me/vsfm/) or AliceVision MeshRoom (https://alicevision.org/). I haven't used MicMac, might be good as well.
I tried several years cheap approaches as you requested with no luck. Then I upscaled my gear.
I've used Agisoft's Metashape with great success (https://www.agisoft.com/). The cheaper license offers really good functionality as is.
You do want to have a good camera for good results. My Mavic Pro's (a drone) 12 MP is barely tolerable. With Sony Alpha 6000 (24MP) and good lens, the results are fantastic. Camera phone can work, depending on the capabilities of the camera, but I would use photography and not video material - the images from photography seem to be better quality than frames extracted from video (YMMV).
If you have the patience to collect the image material, the results can be really good.
For example, as a hobby I've been collecting photogrammetry models from an office building being built near my home:
So what you see there is the model as presented by Sketchfab. The textures and model are coming from Agisoft Metashape, Sketchfab is just used as a platform to display the model for public viewing.
The data closer to ground is captured by my Sony Alpha 6000 while the data from above is from drone. I'm happy with the portions of model based on 24MP DSRL images but the drone based material does loook "melted" occasionally.
As a reference, the source data for that model was roughly 800 images.
The data the photogrammetry algorithms consume are coming from the pixel data. More pixels, the better the outcome. Roughly, the resolution and precision you can expect from the result mesh is equivalent to the pixel density of your source material. The algorithms don't invent anything that they can't see in the pictures. This means, that for example for columns you need to take 360 of tens of images around each column to make them appear ok in the model.
I looked for open source software a while back, but didn't really turn out interesting results :/
Edit: there seems to be a few interesting resources. It looks like MicMac could be quite simple to use, and desn't have a CUDA dependency, for instance.