Beware Rigged China Fever CamerasBy Ethan Ace and John Honovich, Published on Sep 08, 2020
Many China fever camera manufacturers have rigged algorithms dynamically adjusting temperatures to more 'normal' results, IPVM has concluded from 6 months of testing 20+ devices. See the 80-second video below overviewing this:
This raises the risk of missing actual fevers and creates a false sense of safety while producing more 'normal results', since nearly everyone has 'normal' temperatures, even if subjects being measured have characteristics (e.g., eyeglasses, hats, hair bangs, walk in groups or at sharp angles to the device, etc.) that fundamentally undermine thermal temperature screening.
Fever camera manufacturers claim to detect fevers in all sorts of amazing conditions:
This has created false expectations among the public.
Actual / Accurate Thermal Imaging
To the contrary, science, global IEC standards, and US FDA guidelines all have determined the proper and correct way to conduct fever screening with thermal systems is to directly measure and report the temperature captured, as-is:
These images (from the IEC and FDA respectively) are, of course, in sharp contrast to the explosion of deceptive manufacturer marketing.
What Fevers Actually Look Like On Thermal
Accurate temperature screening with thermal systems depends on measuring a tiny section of the face that can be obscured by glasses, hats, hair bangs, etc. In the graphic below, shown in red, we contrast the false expectations that people often have (left) with the small area (on the right) where accurate temperatures can potentially be measured.
Anywhere outside of that tiny area of the face, the temperature is moderate to significantly cooler so readings that miss that tiny area hide actual fevers and produce too low results.
The fascinating element we found in our testing is that these China systems do not produce extremely low results (like 92°F or 94°F or 95°F) even with conditions that fundamentally block actual thermography from getting a good reading.
So how do these China systems get near 'normal' temperatures of a person even when zooming by on a skateboard?
Many China systems we tested rig measurements using "compensation" algorithms to estimate skin temperature. So for example, when the measured max temperature of a person is ~91°F, the camera might add 6 degrees reporting them at 97°, but when measured at 94°, it might add just 4 degrees, reporting them at 98°.
The chart below shows our findings from one of our tests:
This is clever but dangerous. It is clever since these algorithms correctly assume most people have normal temperatures, so they disproportionately increase low readings into more normal ones. And since almost no one has a fever at any given time, it is typically right even if the process is wrong / rigged.
It is dangerous because when the system has an obvious bad reading (like a guy zooming by on a skateboard) it gives a false sense of accuracy. The guy on the skateboard probably does not have a fever since very few have fevers but a thermal system can not determine that directly, since the angle of incident and the speed of the skateboarder makes that impossible.
Missed High Temperatures
The key risk of these algorithms is missed fever temperatures. Because the system corrects readings to more "normal" human temperatures, a subject with a 100°F temperature might be measured as only 98°F.
While these systems will be "accurate" when reporting on people with normal body temperatures (~98°F) and may trigger on those with high fevers, missing low grade elevated temperatures is a systematic problem we found in our testing.
Second, fever cameras display nearly “normal” temperatures on practically any face they detect, so even covering the top half of a person's face (with a piece of cardboard), the camera still provides a measurement close to body temperature:
While the fever camera thought the cardboard was a human forehead, it reported it at 95.2°F, the actual temperature of the cardboard was just 86.4°F.
Indeed, a paper printout of a face is measured at normal body temperature, despite only reading 91° on a calibrated thermometer.
Against Accepted Thermography
This goes against global standards and accepted thermography, as the Chairman of the Board of the American Academy of Thermology declared:
Compensating or self-correcting any temperature would be totally unacceptable. On what basis are they doing that? What do they think they're even measuring temperature if they're just changing the temperature [displayed]?
Multiple People Measured The Same
Finally, these systems produce extremely similar temperature measurements when multiple subjects are measured. For example, in this clip, all three subjects are measured at 99°, even though their temperatures varied .5-by over 1°F according to a calibrated thermometer (97.9-99°F).
Risk: Missed Fevers, Endangering Others
In our testing of 20+ 'fever' devices, we found most China systems show one or more of these issues, failing to accurately detect moderately elevated temperatures. In particular, see our in-depth tests of:
- Dahua Thermal Temperature Monitoring System Tested
- Hikvision Temperature Screening Tested
- Sunell Panda Cam Body Temperature Measurement Camera Tested
These rigged algorithms' favoring 'normal' / statistically average temperatures produce a false sense of safety and raise the risk of infectious people passing inside one's facility.