thermal sensor accuracy

Question

A lot of people are going thermal sensor crazy right now with the application of screening people for fever. I was asked to research it a bit for my company and I two weeks reading papers, googling, and talking to sensor manufacturers. I've found a range of sensors, from expensive $9k-$5k sensors that show 0.1C real temperature data for every pixel, to $200 sensors that have a +/- 5C accuracy. I was sitting here thinking about if there was a way to get better accuracy on the cheaper sensors. My assumption is the +/-5C comes from sensor noise, so it's not like an offset that drifts with ambient temperature, but more of a it might be 5C low this frame and +5C on the next frame. Not sure if I'm right there. But then I've seen sensors with more accurate spot meters in the middle and I was wondering if those can be used to correct for the error? I was thinking about trying to use an external accurate spot meter too but I guess it wouldn't be in sync with the noise.

Anyway my real question is, given a sensor like say the sub $200 FLIR lepton 3.5 who has a radiometric accuracy or +/-5C, is there a technique or circuit design to improve that accuracy? I know component questions are off topic but if there is another type of sensor I should be looking at I'd appreciate any suggestions. For context I have some AR glasses and they can let you "see" in thermal vision using external sensors. And I'm looking for an inexpensive solution that would let them be more widely deployed (as opposed to the $9k camera option).

Thanks!

Answer 1

If you are designing a product using this sensor then the best way to get rid of inaccuracy is calibration.

You will need to use an accurate black body source to calibrate your product. You will need a range of black bodies to use as a reference and design in a calibration routine. It will need some work doing before all this, as you'll need to test the sensor in steps throughout the temperature range to plot a graph to figure out the temperature response. The less linear the response, the more calibration points you'll need. Infrared measurement is a pain to calibrate accurately so you'll need a temperature controlled environment to do it in too.

This can end up getting expensive so it depends how critical accuracy is. But to summarise, the short answer is calibration.

Answer 2

It's more about calibration than noise. Also, Sensors do not have a perfectly linear output and so when they are assumed to be linear as a simplifying assumption, it results in loss of accuracy. The loss of accuracy is greater when you get further away from the calibration temperature. If the sensor is calibrated near to the target temperature, it is possible to get much higher accuracy (though it may drift over a period of a year). If accuracy is most important to you, use a thermistor (though it requires physical contact). See the following links:

The Sensor Shack - IR temperature sensors

The Sensor Shack - Thermistors