Have you looked into OpenSplat type of post-processing? You take a bunch of pictures and then let hardware create a 3d model. It's really competent and could easily create a rectified model for measurements. To get actual values, you'd need some control points, but beyond that, a pipeline that continiously creates models could be feasible.

Then your QC guys are mostly behind computers and rotated to the floor when things are identified.

Ultimately, your VR isn't doing anything more technically accurate than this.