It all started when we have developed a very sensitive heat vision module FLM640 with NETD above 20 mK in the spectral range of 8 — 12 μm, which is quite impressive for an uncooled microbolometer-based camera. The sensor manufacturer contacted us and offered to provide an engineering sample made of experimental plate of bolometer detectors with integrated polarizing filters. It was very flattering for us, but there was no specific end goal in our minds. The technology allowing to visualize self-polarization of thermal photons emitted by regular objects is groundbreaking, and we had no experience processing such data.
In this article we attempt to demonstrate polarization in thermal range, and so far this is the one and only piece on this subject in the Russian Internet that we know of.
We had access to the previously developed FLM640 that guaranteed NETD of the resulting microbolometer ≥ 20 mK. We also had the above-mentioned polarizing sensor the manufacturer provided. Uniqueness of the latter is rooted in the fact that each pixel in every four-pixel group is coated with a linearly polarized film (don't ever ask us how the manufacturer achieved it, we weren't able to extort that information out of them). The four polarizers are oriented at 45° to each other, so their angles are: 0° — 180°, 45° — 225°, 90° — 270°, 135° — 315°.
Processing a dataset from a sensor is no trivial task. If our first attempt was very head-on, the last one had a close resemblance to debayering algorithm (processing of every pixel involves more than four neighboring ones). Sadly, it's worth noting that while the resulting image had thermal resolution of 640 x 512, its polarization resolution is only half as good.
The resulting video presents comparison of three video feeds (left to right): regular thermal camera footage, reconstructed polarization angles, and integrated image where brightness stands for thermal radiation and color — for polarization angles.
In fact, better to see the result once than to read about it a hundred times. That's why we filmed some illustrative videos for this article.
A light bulb
A regular glass light bulb is an excellent object for demonstration of self-polarization. In the spectral range of 8 — 12 μm glass is not transparent and emits thermal radiation that is polarized according to the angle of emission.
A lampshade
This video of plastic lampshade demonstrates how polarization helps visualize the surface structure of objects. If the surface of a smooth object has defects, they can be easily detected this way.
A painted metal container
The way radiation is emitted by flat objects is quite simple: every side has a different polarization angle. In visible or thermal spectral range it would be hard to determine that just by looking at a single frame. Taking polarization into account makes that possible.
A metal plate
A clean metal plate is a complicated object — it doesn't "want" to emit its own radiation, instead it constantly tries to reflect heat from other bodies. The little square in the middle is plastic, and it emits a bit better.
Ice in a cup
The way this looks is quite interesting. In general, polarization highlights surface defects, even some pretty negligible ones. It's possible that it could provide assistance with recognition of cracks in the ice.
UAZ car
And a separate shot of an ordinary street.
Results and conclusions
We managed to demonstrate the most significant aspects of registering objects irradiance in the LWIR range. However, since we are radio electronic equipment developers, not photometry or optics scientists, it's hard for us to estimate the full scope of possible applications of such a device.
For now, we can confidently say that polarization provides some information regarding objects surface structure.
It's safe to assume (on a basis of our interactions with the sensor manufacturer and various colleagues during exhibitions, and also a few scarce internet sources) that estimation of emitting and reflecting objects’ polarization can be used in following areas:
1. Distinction between emitted and reflected light (for example, between a warm car and a sun glare in a puddle, or a rock/sand)
2. Concealed objects detection
3. Oil spills detection
4. Surface defect detection
5. Estimation of objects’ 3D geometry
6. Detection of warm objects (such as a sinking person) on the water surface
Can we increase the sensitivity? Yes, we can, but to do that it's necessary to stabilize the camera temperature (narrow the operating temperature range) and perform some additional calibrations. We haven't done that yes, but potentially it is possible.
