The recent global spread of the novel coronavirus (COVID-19), is changing our lives and capturing headlines around the world. The demand for thermal cameras to screen for elevated body temperatures (EBT) has surged in recent months. The technology is being deployed in airports, businesses and other public places looking for signs of fever, as an indicator of the disease.


Many thermal imaging cameras currently being marketed heavily for fever screening applications are not suitable for this application. In this article, we will attempt to cover the facts and some of the considerations that need to be made if you or your organisation is looking to invest in this technology. 


Thermal imaging can provide an adjunctive temperature measurement of body surface temperatures. It has been effectively deployed to assist with fever screening applications and is likely far more effective than many other non-contact temperature measurement devices. However, there are limitations to the technology. These are limits of bolometer detectors that apply to all makes/models/brands, as well as most handheld temp guns and non-contact thermometers. 


There is a lot of misinformation in the public domain. Many videos of fever screening using thermal imaging cameras are showing up on the internet and on the news. These often show a camera looking at a large number of people flowing through the camera's field of view. These systems would not be providing effective scanning and give a false impression of how thermal imaging can be used. At best this approach may highlight individuals with an elevated skin surface temperature, which is not necessarily a good indication of elevated body temperature. Factors such as if the person is sweating, wearing makeup or moving, can completely invalidate skin surface temperature measurements. 


Following SARS and other infectious disease outbreaks, international standards around fever screening using thermography were developed. If you are planning on using this technology, for this application, within your business, the standards should be used to heavily inform the design of the testing methodology that is put in place - even if you cannot follow the standards to the letter. There are many screening systems being installed, as we speak, that would be completely ineffective and will lead to a false sense of security within an organisation. 


https://www.iso.org/standard/69346.html

IEC 80601-2-59:2017

Medical electrical equipment - Part 2-59: Particular requirements for the basic safety and essential performance of screening thermographs for human febrile temperature screening


https://www.iso.org/standard/69347.html

ISO/TR 13154:2017

Medical electrical equipment - Deployment, implementation and operational guidelines for identifying febrile humans using a screening thermograph


Some of the major points are:

• To get a good indication of internal body temperature a measurement at the inner corner of the eye (canthus) is taken. The canthus is known to provide a reasonable indicative body temperature reading. 
• From a thermography standpoint, to get an accurate measurement you will need around 4x4 pixels covering the canthus. As a guideline, for a 320x240 resolution camera, the face of the individual needs to fill >75% of the image width to ensure enough pixel density for an accurate measurement. 
• The camera needs to look straight on at the face 
• Hats, glasses, sunglasses etc must be removed so that the eyes can be seen clearly 
• A reference blackbody should be included within the frame of the camera, at the same distance from the camera as the face, so as both face/eyes and the blackbody reference panel are in focus at the same time.

These factors can be controlled in an environment such as a passport screening gate or an entry turnstile. However, they are far more difficult to control in an open environment. 

Notes on camera selection

There is a lot of opportunistic marketing going on at the moment. Many sellers are pushing cameras for fever screening that are fundamentally not fit for purpose. Please keep in mind that with any camera, from any manufacturer, marketing doesn't change physics! Neither can fancy software or "AI" for that matter. Anyone saying otherwise should be heavily scrutinised and fact-checked with reputable bodies such as professional thermography associations. Cameras used for "thermal imaging" are built with a different set of design objectives to cameras made for "thermography", i.e. radiometrically calibrated thermal imaging cameras that are designed to measure temperatures rather than just producing a thermal image. 

Let’s run through some thermography basics! 

The instantaneous field of view (iFoV) of a camera is the size of a single-pixel at a given distance from the camera. The measurement field of view (mFoV) of a camera is typically around 4 x iFoV, i.e. 4 x 4 pixels. This is the smallest object that the camera can measure the temperature of accurately. A hot object smaller than this will appear cooler than its actual temperature. 

Note, the mFoV = 4 x iFov specification is for a thermal camera that has great optics. Most thermal cameras do not have great optics.

• Tightly focusing the lens on an object is important for getting an accurate temperature reading. Many cameras designed for security applications or lower-cost cameras designed for thermography applications have a "focus-free" design. This is fantastic for camera ease of use. However, to achieve focus-free operation, compromises in the optical design need to be made. For example, focus-free lenses provide a wide depth of field so that objects close to the camera and far away can all be in "acceptable" focus at the same time. This may require an increase in f-number or similar design changes that reduce the quality of the image that can be achieved. In short, a thermal camera that does not have an adjustable focus lens is probably not suitable for thermographic screening applications. 
• Manufacturing tolerances are not controlled that tightly in many facilities. This results in variation from the  optical properties that the optical engineer intended during the design process, plus further variability between each lens produced. This is often paired with loose quality assurance and calibration standards. A lens that produces an acceptable image to the eye may be perfectly suitable for security applications, but if it is not within tight design parameters then it is likely not suitable for temperature measurement applications. 
• Longwave infrared (LWIR) thermal imaging lenses have traditionally been made of germanium, with high-quality antireflective coatings. Germanium is an expensive raw material and machining costs are high. The adoption of thermal imaging within the security industry has seen a dramatic fall in thermal camera prices. A large part of this has been as a result of the move away from germanium lenses to lenses made of chalcogenide or other similar materials, which can be mass-produced for much lower costs. These materials often don't refract longwave infrared radiation as well as germanium, meaning there can be more dispersion between the 7µm and 14µm wavelengths that the microbolometer is sensitive to. This can make the image focus "soft". The impact from a thermography standpoint is that the requirement for mFoV can become 5 x iFoV or larger. 

Now let’s discuss the application 

Camera Resolution 

First, let’s assume the canthus of the eye is the “hot spot” that we need to measure. The canthus is at most 3mm across as a measurement spot size. Then let's assume that our camera has absolutely perfect optics and we need an mFoV of 3x3 pixels to get an accurate temperature measurement. This gives us an iFoV of 1mm x 1mm, i.e. each pixel is a 1mm high and 1mm wide square at the distance of the face we are measuring. 

Since the mFoV is non-negotiable, this means that the cameras field of view will vary with the resolution of the camera, for example:

• A 640x480 resolution camera can have a field of view no larger than 640mm (horizontal) by 480mm (vertical). 
• A 320x240 resolution camera can have a field of view no larger than 320mm (horizontal) by 240mm (vertical). 
• A 256x192 resolution camera can have a field of view no larger than 256mm (horizontal) by 192mm (vertical). 
• Etc. 

Keep in mind that the above is a best-case scenario, assuming absolutely perfect optics, which will not be the case in the real world. This is the reason that accurately screening multiple people from across a room is just not possible, irrespective of the amount of snake oil in the sales pitch. As a minimum, a camera with 320x240 resolution and focusable optics should be used. The face of the person being screened should fill most of the camera's field of view.  

This aligns with the statements buried deep in the user manual for cameras like the FLIR A320 Tempscreen that states “The distance to the face should be adapted so that the face covers more than 75% of the image width”, even if this is not the workflow that is portrayed in many videos and photos of these cameras in use. 

Camera measurement accuracy 

The same rules that apply to standard thermography applications apply to fever screening. It is very easy to enter incorrect values for emissivity, or other settable parameters, into a camera. These will make the camera return absolute nonsense temperature measurement results. There is lots of information available that explains thermography basics. 

https://www.flir.com.au/discover/professional-tools/how-does-emissivity-affect-thermal-imaging/

Sticking to the fever screening application, there are two approaches to temperature measurement that are used, each addresses the error in measurement accuracy of the camera in a different way:

• Absolute temperature threshold, e.g. a measurement of >37.4°C detected 
• The rolling average method, e.g. an individual’s temperature is >1°C higher than the average of the last 10 people screened

Both have different camera requirements and configurations, which are discussed separately below. 

The absolute temperature threshold method 

Most thermal cameras will have an absolute measurement accuracy stated as something like “the greater of ±2°C / ±2% of reading”. When we are talking about an EBT tolerance of around 1°C, this accuracy is not sufficient. To utilise these cameras we can place a blackbody reference panel within the field of view of the camera. This blackbody is a near-perfect emitter of thermal energy that the camera can measure the temperature of very accurately. 

Camera focus is important to temperature measurement accuracy. It is important to make sure that the blackbody is at the same distance from the camera as the face of the individual being screened. This ensures that both the face and the blackbody can be in sharp focus at the same time. It is also important that the blackbody reference panel is large enough for the camera to measure very accurately, a size larger than 10x10 pixels should be used as a minimum guideline. The measured temperature of the blackbody can then be used to provide a temperature correction offset to the camera. If this approach is taken then the measurement accuracy of the camera will be in the order of 5x-10x the NETD specification.

Thermal sensitivity or Noise Equivalent Temperature Difference (NETD) describes the smallest temperature difference you can see with the camera. The lower the number, the better the thermal sensitivity of the infrared system. 


https://www.flir.com.au/support-center/Instruments/how-is-nedt-measured/


Thermal cameras designed for security applications often have a NETD as high as 150mK. To be suitable for a screening application the NETD of the camera should be no greater than 50mK. Even if a camera with 100mK NETD is used with a blackbody reference panel, the measurement error will be around 10x100mK = 1°C, which is before we factor for any variability in the blackbody itself. This is not suitable for screening applications. It is also worth noting that the way this is calculated varies between manufacturers. Cameras from low-cost manufacturers may be hiding poor sensitivity by taking NETD at 50°C instead of the industry-standard 30°C.

There are some other considerations that need to be factored in. 

• Linearity across the detector is important. Thermography cameras correct themselves with a non-uniformity correction (NUC) shutter. 
  
  
  https://www.flir.com.au/discover/professional-tools/what-is-a-non-uniformity-correction-nuc/
  
  
  Cameras designed for security applications minimise the number of NUC cycles, as this causes the image to freeze for a few seconds each time it occurs, which is not ideal if that is the moment when the bad guys are sneaking past the guard. However, for temperature measurement applications it is important that NUC cycles are performed regularly to optimise linearity of the image and minimise measurement drift over time. 
  
  If the temperature of a blackbody is tracked between NUC cycles there will be some drift in the measurement over time, this will result in a "sawtooth" pattern in the temperature measurement. With a high-quality thermography camera, the peak-to-peak change in this sawtooth will be <0.1°C. With a camera that is optimised for security applications with "radiometry" functionality added on, the peak-to-peak sensor drift over time can be >1°C. This is not suitable for screening applications where repeatability across individuals is critical. 
  
  As a thermography camera is coming up to steady-state operating temperature it will NUC more frequently, it is recommended that for screening applications the camera is running for around 30 minutes before screening commences. 

• Optical artefacts, such as vignetting, can cause temperatures at the edge of the frame to differ from temperature measurements at the centre of the frame. Good quality thermal camera optics should minimise this, but cameras designed for applications where some vignetting does not matter (like cameras for security purposes) often have very bad optical artefacts present and can measure >2°C different between the centre of the image and a corner of the image. This can be problematic if the blackbody is placed at the edge of the image and the face in the centre. 
  
  
  
  

Notes on accuracy specifications stated on datasheets (added 2020-03-30)


In recent weeks there have been numerous thermal imaging systems entering the market targeted at COVID-19 screening applications. Many of these cameras have extremely high accuracy stated on their datasheets, for example, ±0.1°C. These types of statements raise alarm bells for most professional thermographers.


The way that some manufacturers are justifying these accuracy claims is by using the NETD of the camera, as the accuracy, for cases where an in-frame blackbody reference panel is present. This is not correct for several reasons. For example, NETD is calculated using an average of all pixels and may not be representative of any individual pixel, including the pixels that are located over the target being measured. It also assumes that the blackbody reference panel has perfect stability and no temperature uncertainty, this is not the case in the real world.


This highlights that there are significant differences in the way that manufacturers calculate the specifications that are provided on datasheets. The calibration of a thermal imaging camera is a complex process. This article explains some of the intricacies in more detail.


https://www.flir.com.au/discover/rd-science/infrared-camera-accuracy-and-uncertainty-in-plain-language/


It is worth highlighting that the calibration of the camera varies with the ambient temperature and the temperature of the sensor and optics. High-quality thermography cameras are calibrated across their operating temperature range to ensure the calibration is valid under all operating conditions. Many cameras are not and will be calibrated at a single operating temperature only. The calibration certificate provided with your camera should state the testing conditions and limitations for the calibration. It should reference traceable standards (NIST etc) for any radiation sources used during calibration.


IEC 80601-2-59 provides very clear guidelines for how the accuracy is to be assessed and verified. The system vendor should be able to provide guidance on this.



Example system configuration

Camera specifications:

• IR resolution: 640 × 480 pixels
• Thermal sensitivity/NETD: < 0.03°C @ +30°C / 30 mK
• Field of view (FOV): 25° × 19° (31° diagonal)
• Spatial resolution (IFOV): 0.68 mrad
• Object temperature range: –40°C to +150°C 
• Accuracy: ±2°C or ±2% of reading

Backbody reference panel specification:

• Target Temperature: 35 °C
• Emissivity (ε): 0.98 ± 0.004 (for wavelength of 8µ to 14 μm)
• Aperture diameter: 80 mm
• Temperature uncertainty: 0.4 °C for T(ambient) = 10 to 30 °C
• Repeatability: 0.2 °C 
• Stability: 0.1 °C 
• Temperature uniformity: 0.2 °C 

How the screening system is configured:

• We mount the camera at head height and mark the floor where people can stand 1.4m in front of the camera. This gives us an mFoV of 2.9mm and a horizontal field of view 61cm wide with the lens stated. 
• We mount the black body reference panel on a stand next to where the individual's face will be at 1.4m distance from the camera. The blackbody is pointed directly at the camera.
• A measurement spot within the camera software is placed in the centre of the blackbody reference panel and appropriate local parameters adjusted (emissivity, distance etc). 
• A measurement area box covering the area of the face is configured. 
• Individuals are asked to step up to the screening location one by one, remove hats, glasses, goggles etc and then stand still while an image is captured. This is fast, less than a second required.  
• The camera logic is then configured something like:   
• ∆T = T(Area Max) - T(Spot) 
• Alarm if ∆T > 2.4°C 

This will alarm on the condition that any person’s temperature measurement is above 37.4°C. A temperature of 37.5°C at the canthus is typically recommended as a good point to start screening for potential fever. The error in the measurement is <0.7°C (10x 30mK NETD + 0.4°C BB uncertainty), with a good quality camera, lens and blackbody the error should be significantly less than this. 


It is important to understand that all the camera is indicating is the presence of elevated body temperature (EBT). The camera is not diagnosing an individual as having a fever, COVID-19 or any other medical condition. Once the system indicates an EBT, procedures need to be put in place to allow an appropriately qualified and licensed medical practitioner to assess the individual. 

How to use the “Rolling Average” average screening method 

All of the same requirements regarding iFoV, mFoV and working distance discussed above are valid to the rolling average method. The most important factor, particularly if using a handheld camera, is consistency, i.e distance to the person, measurement parameters etc.  

The easiest to deploy solution for Elevated Body Temperature monitoring is to use one of FLIR’s handheld cameras in “Screening mode”.  Below is a link to some short videos showing how this works.

www.flir.com/ebt  

https://www.youtube.com/watch?v=GVWZ-70Pkds&feature=youtu.be.  

https://www.youtube.com/watch?v=GVWZ-70Pkds&feature=youtu.be

Software that runs on a Windows based computer is also available that can perform the rolling average screening method with any of the FLIR handhelds or mounted (A-series) cameras. The cameras should meet the minimum specification guidance of 320x240 pixel resolution with focusable optics.  

Summary 

Thermal imaging technology is brilliant. It has many fantastic applications and can be used effectively for Elevated Body Temperature (EBT) screening applications. However, it is important to understand the limitations of the technology. Thermal imaging cameras produce pretty pictures that make it very easy to be lulled into a false sense of security that all is ok. There are a lot of thermal cameras that were never designed for accurate temperature measurement, which are being hastily rebranded and marketed as a “coronavirus detection camera”. No camera can detect coronavirus. The camera can only measure surface temperature differences, which can be an indication of elevated body temperature and issue with that person. Only a licensed medical professional can determine if a “hot” individual is experiencing an abnormal medical condition. 

This technology has the potential to be of great assistance during this period of crisis. But please do not buy into the spin that it is a silver bullet, it is a very small part of a very big solution, to an enormous problem that we are all facing together.