[nitrogen]

You'd be looking at very subtle variations in the color of single pixels, so you might as well just use a single optical sensor and light source into a single high res adc, using intensity to measure thickness.

At some point dust in the air will be a limiting factor.

[elcritch]

For high accuracy, the brightness is really more the limiting factor, as most of the edge finding/binarization algorithms use a intensity factor for cutoff. I used camera measurement for small pieces of tissue samples in biomechanical measurement (~1mm x 5mm). We had a small window in our lab, and cloudy days would distort measurements. I solved it mostly by going to a SLDR camera from a handheld and a backdrop with a type of luminance chart. Much more accurate than calipers for soft squishy (and wet) samples. See [1] Figure 2E (pg 22), you can average out small variations pretty readily.

1: https://pdf.sciencedirectassets.com/274145/1-s2.0-S175161611...

Though your idea of a light source and adc could work well if you put it in a darkened enclosure. It’d probably require a bit more hardware design or tuning but the software / microcontroller side would b simpler. A RPi PiCam with a macro lens would work well too, but the reverse with less hardware tweaking and more software.

[londons_explore]

You'd make the filament go slightly diagonally across the field of view of the camera, and then you'd have thousands of part-filled pixels.

As long as your material is consistent (no bubbles, shiny spots, etc.), You should be able to build a very accurate sensor this way.

[econcon]

I found this video https://youtu.be/kpsN78mQ6YY

They use a laser pointer to measure width of a hair.