Hokay, another look at this.
Briefly, the
Fresnel Equations tell us that the reflections from an air-glass interface (or any two transmissive mediums) will have a degree of
polarisation related to the
angle of incidence. In plain English this means that if you shine unpolarised light at a windscreen then the reflected light will be polarised one way more than the other, depending on the various angles. This works at any wavelength, LWIR, ultraviolet, whatever. If you use a polarising filter on a camera (or eyeball) looking at the reflection, then you will be able to find a position where the filter matches the reflection best (you'll see more reflected light) or the transmitted light best (you'll see what's behind the glass).
Conversely, if you shine polarised light onto the windshield, you'll get more or less transmission vs reflection as you change the polarisation of the light source.
Sunlight is only very slightly polarised (this is down to atmospheric effects and both the polarisation angle and amount of polarisation vary during the day). So it's likely that at I could adjust a camera and polarising filter for minimum reflection at (say) 9am but by later in the morning it would require readjustment for optimum results. That's why I mentioned an array of cameras in an earlier post - one optimised for (say) each couple of hours.
At night things are different, as one can have control of the polarisation of the light. You know the likely angle (or at least orientation) of the car windscreen so you can arrange the polarisation for maximum transmission and minimum reflection. You can also place the emitter(s) where they won't be
directly reflected from the glass. Furthermore, you can put a suitable polarising filter on the camera so that any light scattered into the 'wrong' polarisation will be rejected.
As a side issue, I believe that using relatively narrow band light (eg from a LED source) makes all of this easier still, as you're then dealing with a much smaller band of frequencies and can optimise the optical elements accordingly.
The bottom line is that if you have reasonable control of the environment it's not too hard to arrange near-IR lighting to give excellent see-through-glass capability, at least most of the time. Daylight - which contains a lot of near-IR - complicates matters.
It's also worth contemplating that one only releases one's best shots for publicity purposes. We have no way of knowing whether 99% of the images are as good as the samples shown, or if they were one-in-a-million lucky shots.
Moving back to longer wavelengths, has anyone got any experience with polarising filters for LWIR? Do they even exist?