Detecting Coronavirus Fevers With Thermal Cameras

Published Mar 15, 2020 05:53 AM
PUBLIC - This article does not require an IPVM subscription. Feel free to share.

MAY 2020 Update: This post was our early examination of these systems being used in the pandemic.

For more up to date coverage, see The Booming Multi-Billion Coronavirus Fever Camera Market and World IEC Fever Screening Standards Explained as well as our growing library of tests:

Demand for thermal cameras as a tool for authorities to help detect fevers has skyrocketed as coronavirus spreads across the world.

IPVM Image

But there are significant concerns about how accurately they work. In this report, we examine:

  • Biggest accuracy issues
  • Setup examples
  • Problems examples
  • Temperature variance issues
  • Optimizing accuracy
  • Survey of options ranging from FLIR and Optotherm from the US to various China suppliers including Dahua, Hikvision and Sunell
  • Performance claims and pricing differentials for thermal products offered.
  • The rising importance

Biggest Accuracy Issues

In our research of various options, we found 4 fundamental accuracy issues:

  • Camera temperature detection accuracy - while some manufacturers claim accuracy to as precise as 0.3°C, we are skeptical of actual field accuracy as these are likely overinflated. Surveillance users can compare to lux ratings.
  • The positioning of cameras - many cameras being used to detect temperature are positioned perpendicularly to the person and/or the person is on the move. This significantly reduces the probability of an accurate reading, as one is briefly getting the side of a person's head, which typically has a lower temperature reading.
  • Reading temperature on the forehead and other areas of the face is heavily impacted by environmental factors. In airports, for example, subjects entering from outside will be skewed by hot, sunny days or very cold temperatures, or those running to catch a flight may be flushed, raising surface skin temperature.

  • The use of blackbody devices (powered heated elements) is rarely emphasized but frequently required by such systems, increasing the cost and complexity of deploying them.

Pricing variance: Beyond accuracy issues, the other major issue that stood out was drastic differences in pricing, ranging from the ~$10,000 range for US cameras like FLIR, Optotherm and some China systems that are marketed specifically for human body temperature reading to hundreds of dollars for China models. However, FLIR does offer much less expensive cameras that detect temperature but do not recommend because of accuracy limitations.

Key Issue: No Independent Tests

A core issue is there are no independent tests of thermal camera performance/accuracy and no independent standards to measure against. This has allowed manufacturers to tout products meant for body/fire detection as a fever solution, or falsely claim pinpoint accuracy at long distances. We urge caution against buying low-cost thermal solutions from any manufacturer.

UPDATE May 2020

IPVM has now conducted a series of independent tests, including:

IPVM is continuing to publish new test reports.

Setup Examples

First is an example from Wuhan, the origin of the epidemic, where a NY Times reporter took a photo of herself having her temperature detected. We have translated the screen for context:

IPVM Image

Despite translating and searching in both English and Chinese, we could not find the name of the manufacturer. The on-screen text indicates it the device combines color and thermal cameras, color for display purposes and thermal for measuring temperature.

In some cases, people are moving directly towards the camera, as this AFP photo shows FLIR in Malaysia:

IPVM Image

Here is another case where the person is moving directly but notice how the person head often will turn the other way, as this AFP news video of the airport in Hanoi, Vietnam using OptoTherm equipment can be seen:

IPVM Image

In the case below, the camera (highlighted in the red box) is at a moderate angle to traffic, as shown in this NY Times report showing a Thailand mall:

IPVM Image

The crowd of people makes it more difficult to tell. Note; red is for alarming on temperatures over 38.0° C ( 100.4° F).

By contrast, other detectors are used to measure temperature person by person as this example in the Philippines shows:

IPVM Image

Of course, the obvious tradeoff of this vs thermal cameras is that a person is required to measure temperatures one at a time, which requires a lot more time and staff to implement.

However, the plus side of having a person read temperatures one by one is decreased errors due to people's heads not being properly aligned with cameras and having the sensor much closer to the person.

Problems Examples

A common problem is that readings come in too low, especially when the camera is reading the side of someone's head:

IPVM Image

And again in the example below:

IPVM Image

There are also examples of a handheld infrared thermometer having significant accuracy issues, as shown by this Economist reporter who tweeted that one gave him a 32.1ºC / 89.8ºF temperature, which would mean he is dying of hypothermia:

IPVM Image

Neither the model nor make of that device nor other particulars were available. The underlying point of under readings, though, remains clear.

Another example of infrared thermometers' unreliability comes from an NYT reporter, who tweeted he was constantly getting "corpse-level" 34ºC readings:

IPVM Image

Skin Temperature Lower Than Body Temperature

Likely contributing to these issues is that skin temperature is colder than someone's actual (internal) body temperature. Known as the "skin temperature offset" as explained by OptoTherm, this difference can be a few degrees Celsius:

IPVM Image

Optimizing Accuracy

IPVM Image

To get the most accurate readings, according to a source in this field, one needs:

the person face the camera, remove glasses and allow the camera to image the inner canthus [corner] of the eye

Glasses can distort the readings since they interfere with measuring heat from the body. Moreover, the inner canthus of the eye emits higher, more accurate temperature than the skin.

Moreover, like with cameras generally, the pixel density has a factor with lower pixel density increasing accuracy issues:

Trying to get a reading of a person far away will also significantly decrease performance as recommendations are to get 9 to 16 pixels on the 1/4 inch area of the eye, which effectively limits the overall camera's FoV

One technique that is sometimes used is effectively doing relative measurements, such that what stands out is a person whose readings are higher than others in the immediate vicinity / time frame. This still has the risk of people with fevers being missed if the camera only reads the side of their head at a poor angle or distance, etc.

Blackbody UsageIPVM Image

Many of these systems use a blackbody, an expensive device that emits heat at precise levels to help improve the accuracy of the thermal camera measuring people's temperatures.

The problem is that the difference between normal human temperatures (say 37°C) and fevers (say 38°C) is very small (just 2.7%) so even slight errors in measuring temperatures can cause false readings.

The goal of the blackbody is to provide a fixed, accurate temperature reading so that the thermal temperature measurement system can reduce their error rate.

The blackbody is positioned in the thermal camera's Field of View as this excerpt of a Dahua video shows on the left side:

IPVM Image

And below is an example layout from Uniview showing a blackbody deployed in a similar fashion:

IPVM Image

Finally, below is an example from a Hikvision marketing demo showing its blackbody in the FoV of the thermal camera:

IPVM Image

The blackbody needs to be powered and it needs to be mounted close to, if not somewhat in the way of people walking by. It also needs to be calibrated and work correctly or else it will make the system less accurate.

Accuracy

Accuracy claims are difficult to assess as some manufacturers are conservative and others almost certainly lying. For example, FLIR states that thermal cameras typically have accuracy +/- 2ºC, but that "with proper calibration and attention to factors such as ambient temperature, emissivity, and spot size, the possible margin of error can be less than 1ºC." Other manufacturers, particularly in China, have claimed up to 0.1°C accuracy, though, of course, whether they are accurate in their accuracy claims should be treated with skepticism.

IPVM Image

If deployed properly - with a focus on individual people rather than large groups all at once - thermal cameras are useful, but will not catch all those infected, said Ben Cowling, a professor at the University of Hong Kong's School of Public Health:

Yes quite a number of the imported cases have been picked up by thermal scanners. It can pick up people with a high body temperature. Those can then be verified with a hand-held thermometer. It does not pick up people who have been infected but do not yet show symptoms ("during the incubation period"). For the new coronavirus, the incubation period could be up to 14 days. So it will catch some infected people but miss others. It does not pick up infected people with symptoms but who do not have a fever, for example the Chinese tourist who went to France, departed with a fever but took paracetamol to evade the thermal scanners. Some infected people may have symptoms (e.g. coughing) but no fever. It is useful but will not identify every infected person.

Limited Studies

A 2010 study from the US' Center for Disease Control that a FLIR A20M camera and an Optotherm Thermoscreen were each "reasonably accurate in detecting fever and were better predictors of fever than self report". IPVM could not find a more recent study as authoritative.

However, worth noting the test was limited to the much more expensive units and is a decade old.

FLIR Offerings

While FLIR has various thermal cameras, FLIR only markets a limited number of thermal cameras for body temperature detection. Notably, FLIR does not recommend its conventional security thermal cameras for body temperature detection.

FLIR provided us a list of models they say may be considered include the portable FLIR E75, E85, E95, T530, T540, T840, T860 and the FLIR T1020 (T1K), or fixed mount A310, A315, A615, and A655sc, the latter of which is a more accurate, higher resolution, but more expensive option.

US FDA 510K Certification

In the US, there is a Food and Drug Administration (FDA) certification called 510(K) which impacts the use of thermal for detecting body temperatures, according to both FLIR and Optotherm.

FLIR says:

These FDA-certified cameras (510(k) submission K033967) were designated with an intended use specifically in public areas, i.e., airports, to allow operators to visualize and document temperature patterns and changes.

Here is the FDA 501(K) listing for FLIR and the 2004 summary PDF report.

We do not understand enough how the 501(K) process works or how it actually verifies performance since it requires comparison to devices released prior to 1976:

Submitters must compare their device to one or more similar legally marketed devices and make and support their substantial equivalence claims. A legally marketed device is a device that was legally marketed prior to May 28, 1976

OptoTherm Offering

OptoTherm sells a single solution called Thermoscreen. This clip from the AFP news video shows it in action:

Notably, OptoTherm declares on their website that it cannot be sold in the US due to a current lack of 510(K) certification:

This product is currently for export only. Optotherm is currently undergoing the US FDA Premarket Notification (510k) submission process for medical devices. As soon as Thermoscreen has been cleared, this product will be available for purchase in the US.

PRC Manufacturer Dubious Claims

IPVM has noticed several PRC manufacturers touting their thermal solutions in light of the Wuhan virus. However all of them had significant issues:

  • A Sunell thermal camera demo showing pinpoint accuracy at a long distance, which is suspicious to impossible
  • Dahua claiming its thermal cameras can detect fevers with pinpoint accuracy, even though they are meant for detecting fires in electrical substations and bodies at border crossings
  • Hikvision deleting a tweet showing a wide range of body temperature detections for a single person

Panda Cam Claim from Sunell

A Sunell employee on LinkedIn posted that their "Panda" camera could help "protect us from Coronary virus" [sic]. This is the Sunell TN-5 we saw at ISC West last year.

IPVM Image

However, this solution is unproven and makes a number of dubious claims, including pinpoint (to a fraction of a degree) temperature readings for people at a significant distance, as shown in this demo:

IPVM Image

At such a high distance, with little to no focus on the crucial inner part of the eye, such a specific body temperature reading is simply not possible.

Even more astounding, this demo gets 4 readings in a row all within a fraction of 1 degree Celsius from ideal body temperature, which means this is astoundingly lucky, incredibly accurate or simply rigged:

IPVM Image

Additionally, in its specs, Sunell states the Panda cam has +/- "less than 0.3°C" accuracy in ideal conditions, which is far more accurate than the typical +/- 1°C accuracy in ideal conditions that FLIR says is possible. Their specs also don't include a recommended minimum distance or a description of what ideal conditions actually are.

IPVM Image

There is no evidence the Panda camera is being deployed against the coronavirus. When IPVM reached out to Sunell, they were unable to answer any questions about the SN-T5.

Hikvision Deletes Thermal Tweet, Sends Equipment to Wuhan

Shortly after the virus became international news, Hikvision Italy's Twitter account shared a video of thermal camera surveillance footage referencing current events, implying that Hikvision products could be used as a coronavirus solution. Below is the video, which shows a considerable range of temperature:

However, when we reached out to Hikvision, they deleted the video and told us:

We have deleted the tweet, it was part of an ongoing thermal campaign in Italy and could now be deemed misplaced.

Separately, a local news report from Hikvision's Hangzhou HQ said that on January 22, Hikvision dispatched nearly 1,000 "sets of video capturing and analytics products as well as thermal detection devices" to Wuhan's Hospital Number 7, sending another 40 sets of thermal devices a day later due to local shortages:

IPVM Image

IPVM Image

The specific Hikvision devices being packaged are the H10 infrared thermometers, which are sold on Alibaba for about $288 (1999 CNY) and also have reported accuracy of +/- 2ºC. This is the same accuracy as FLIR, a company that specializes in thermal solutions, and we have no way of confirming whether this reported accuracy is true. These devices are also only advertised for industrial purposes.

Another news report states Hikvision has set up an emergency committee to deal with the virus response.

Hikvision declined to comment on its response to the Wuhan virus.

Dahua Touts Unrelated Thermal Product Line

In response to the virus, Dahua's Chinese social media account issued a post titled "Dahua will always guard you!" that touted its thermal camera and access control solutions, claiming accuracy within 0.3ºC like Sunell.

IPVM Image

However, 3 of the products above have different purposes than human body temperature, according to Dahua's own promotional video.

The two bullet cams (BF5x00 left, BF2120 right) are meant for indoor monitoring/perimeter protection/fire detection while the camera on the lower right (PT8x20b) is meant for city surveillance (i.e. body detection) and detecting fire hazards at electrical substations. We could not identify the fourth camera. Furthermore, while Dahua's post touted +/- 0.3°C accuracy, these cameras' specs all list +/- 2°C accuracy.

Fire and body detection, obviously, requires much lower accuracy than detecting a human fever, but there is no mention of these cameras' original purpose in Dahua's post. The promotion of this series amid the coronavirus crisis showcases the danger in taking surveillance manufacturers' claims at face value.

PRC Government Mandates Anti-Virus Equipment Production

China is currently experiencing critical shortages of thermal detection equipment, face masks, hand sanitizer, etc, leading the PRC government to announce measures ensuring the production of such goods:

IPVM Image

Although this directive mandates enterprises to "strengthen product quality management", IPVM has not found any mention here (or anywhere else) of specific measures targeting thermal screening accuracy.

Delivery Food Includes Temperature of Those Involved

The measuring of temperatures is becoming a broad phenomenon inside of China. For example, this Twitter user shared an image of a delivery slip where the "food preparer, packer, and the courier all have to measure their temperatures":

IPVM Image

The card above is from Yunhaiyao, a chain restaurant serving Yunnan cuisine across major cities in China and overseas. We saw other reports on Weibo of similar activities from other chains.

Outlook - Thermal Challenges But Difficult Conditions

While using thermal cameras for temperature detection presents challenges, the difficult conditions faced clearly have spurred their use. And the quick acceleration of these challenges means that most users and few suppliers are ready with highly accurate thermal detectors.

We plan to update this report as we have more information on accuracy, best practices, options and usage of thermal cameras detecting body temperatures and alerting on fevers.

[Note: This post was originally published in February 2020 when thermal camera use first emerged in China and updated in March 2020 as coronavirus has spread worldwide.]

Comments are shown for subscribers only. Login or Join