How thermal imaging cameras work As far as I see from wiki, 'consumer'-grade(non-cryogenic) thermal imaging cameras use microbolometer sensors to get integrated IR intensity over some 5-12um range. But I had an impression, that having just integrated IR intensity you can't get surface temperature precisely, especially with 0.1C precision we see on consumer cameras.
IR thermometers (pyrometers) for example use 2-band IR measurement, but I don't see this principle being used in thermal imaging. 
Or there is some sort of 'know-how' matrix of IR band filters in front of the sensor?
 A: I'm not an expert at thermography per se, but I've designed and built a commercial microbolometer IR camera and looked at several other cameras (for thermography) and as far as I know, they don't have any "funny business" going on. Commercial handheld thermography cameras require you to input the surface emissivity, air temperature and transmittance parameters etc to get an accurate reading, so obviously in practice for random measurements you won't get anywhere near +/- 1K accuracy, but the noise (precision) can be as low as dozens of mK like quoted.
In industrial (repeatable) applications you can probably calibrate this very accurately though.
A: Here  http://www.jenoptik.com/en_40150_thermographic_cameras
they claim +/- 1.5K accuracy.
In this spec another company claims 50mK Minimum Discernable Temperature Difference:
http://www.argusdirect.com/assets/media/files/datasheets/cameras/argussc-specialist-search-camera/argus%20sc%20english.pdf
The precision of these bolometers is nice but the accuracy is more than 10 times worse:
http://en.wikipedia.org/wiki/Accuracy_and_precision
Unfortunately I couldn't find any datasheets on the sensors on the e2v site.
