Infrared Training Center

Thursday, February 23, 2012

Accuracy Specification of FLIR Cameras

How accurately can the camera measure an absolute temperature?

Most FLIR Thermography cameras have a specified accuracy of  ±2 ºC (±3.6 ºF) or ±2% (whichever is greater) of reading for a blackbody target (emissivity ~ 1).

For example, for objects that are 100 °C or lower, the temperature reading off a blackbody can be 98°C to 102°C and be within specification. Similarly for objects above 100°C, say 200 °C, the reading could vary between 196°C and 204°C.

Some science cameras such as the SC660 are specified ±1°C or ±1% of reading.

This means that any camera, at any environment condition (within specification), at any time will give a reading within the accuracy specification.

However, a particular camera, at the same environment condition, will have a statistical repeatability of measurement that is much better than this. Typically close to the NETD value. This is also applicable when you compare adjacent pixels, provided your target(s) are optically resolved. This means that much higher accuracies can be achieved by comparing values with a known reference source in the image scene.

What happens when measuring real objects with emissivities less than one?

When measuring materials with emissivities less than one, in other words real everyday objects, the measurement error is increased. A simple to apply rule of thumb is shown below:

                                      Camera accuracy specification
Adjusted accuracy =  --------------------------------
                                  surface emissivity

So let’s take our 100°C object again, but this time the emissivity is 0.7. So the adjusted accuracy would be approximately:

  ±2 ºC (±3.6 ºF)
Adjusted accuracy = --------------------- = ±2.9ºC (±5.1ºF)

  0.7

For specific applications such as fever screening, there is an averaging function implemented that makes absolute accuracy specification unnecessary. The averaging function sets a reference value from a series of "normal" readings, and gives an alarm if one single reading differs more than a certain limit.

4 comments:

  1. Hi Gary,

    This will really help us take the most accurate measuremants possible, thank you! It was interesting to see how to take emmisivity into account too.

    I have a further follow up:
    We are trying to make some calculations about the distribution of the error. Suppose we have a blackbody object of emissivity ~ 1. And further suppose the Flir Infracam or Infracam SD is within specification.

    1. If we take multiple pictures with the same camera (in a short time span), will the mean temperature be the absolute temperature?

    2. Suppose we we took multiple images of the same object with the same object, and the mean temperature is M (now, regardless of whether this is the absolute temperature or not). If we look at the distribution of the termperatures from each image, will the errors fall +-2% or +-2 C of M 98% of the time? And will the distribution around the mean be a Normal distribution? Or something like this shape:
    http://www.risk-management-basics.com/images/big/Probability_density_function_PDF.png

    or neither?

    -Robert Simione

    ReplyDelete
    Replies
    1. In order to address your questions, I will need details on your application and use of the camera.

      Delete
  2. Gary,
    I'm concerned with object temperatures only 5 to 20 C above air temp.

    I'm using an e=1 paint spot to determine the object temperature and emissivity. The real emissivity will be unknown, but between 0.4 and 0.9.

    If the temp accuracy is +/-2C, do you how to relate it to +/- emissivity accuracy?

    Dan Lawry

    ReplyDelete
    Replies
    1. From my calculations, the maximum error you would get on a 43 C target (23 C ambient) with an emissivity of 0.4 would be about +/- 3.3 C.

      Delete