Glare is the most common reason a vision system "sees nothing" on a part that looks perfectly clear to a human standing at the same station. The camera is not broken and the lighting is not dim - the light is simply arriving at the lens after a mirror-like bounce off the surface, carrying almost no information about the surface itself. Polarization is the cleanest physical fix for that problem, and it is also one of the most over-applied.
What a polarizer actually changes
Light reflected off a smooth, non-metallic surface tends to become partially polarized in the plane of that surface. Specular glare is therefore not identical to the diffuse light coming back from the bulk material: it has a preferred polarization direction. Put a linear polarizing filter on the light source, another on the lens, and rotate the lens filter until the two are crossed - and most of the specular component disappears while the diffuse component survives. The effect is physical, not algorithmic, so it works before noise, exposure or resolution ever enter the discussion.
Where crossed polarizers are the right answer
Shrink wrap and blister packs. Printed codes and lot text under glossy film are effectively invisible to an unpolarized setup. Crossed polarizers turn the film transparent to the camera without changing the product.
Bare metal and machined surfaces. Cut edges, chamfers and turned surfaces produce long specular streaks that saturate whole rows of pixels. Crossed polarizers suppress the streaks and let tool marks, scratches and burrs appear at their true contrast.
Glass and glossy molded parts. Inspecting for chips, cracks or inclusions in glass is otherwise a fight against window reflections from the ceiling or from the machine frame itself.
Liquid surfaces and filled containers. Fill level and presence checks on liquid become stable once the meniscus highlight stops dominating the region of interest.
The price you pay
A crossed pair of polarizers typically removes two to three f-stops of light. That has to come from somewhere: longer exposure, brighter or pulsed illumination, or a larger-aperture lens. On a moving line, longer exposure is usually the option you cannot take, so polarization is normally paired with strobed LED illumination rather than with a continuous light.
Second, polarization is orientation-dependent. A part that rotates freely on the conveyor presents a different surface orientation every cycle, and a fixed filter angle will only be optimal for some of them. If orientation is uncontrolled, a dome light or a diffuse coaxial arrangement usually beats polarization on repeatability, even though it fixes less glare on any single frame.
Third, polarization does not help at all with diffuse scattering from rough, matte surfaces - there the reflected light has already lost its preferred direction, and the filter only costs you signal.
Building a practical polarized station
Mount the polarizer on the light first, then on the lens, then rotate only the lens filter while watching a live histogram. Stop where the glare peak collapses but the part's own contrast is still climbing - the two optima rarely coincide, and the right setting is the one that maximizes the contrast of the defect you are trying to detect, not the one that looks darkest.
For a mono GigE station the rest of the chain is unremarkable: a 5 MP sensor such as the MV-CS050-10GM at 3.45 um pixels gives you enough signal headroom to absorb the two stop loss, and a low-distortion fixed focal lens such as the MVL-HF2528M-6MPE keeps the polarized image geometrically intact. If you need higher resolution across the same field of view, the MV-CS200-10GM at 20 MP leaves more pixels on the defect after the light loss.
The summary worth keeping: reach for polarization when glare is specular, the surface is smooth, and the part is held in a repeatable orientation. Reach for better geometry - dome, coaxial or dark-field - when any of those three conditions fails.