For years, thermal-imaging cameras have been used to enhance video surveillance. Where most cameras and human eyes see nothing, these see by translating thermal energy (heat) into visible light. In other words, they can see the heat radiating off an object by capturing different levels of infrared light. They are ideal for screening the perimeter of a property at night.
In recent weeks, there has been a significant increase in the demand for thermal cameras. This rise in interest correlates to a time when governments and businesses are looking for quick and scalable ways to track who is sick. Their thinking is that thermal cameras can pick out those people who have elevated temperatures, a symptom of COVID-19, to help combat its spread.
Seems like a logical idea. However, experts say thermal cameras are an imprecise tool for identifying people who may be infected with the virus.
A recent article published by IVPM, the leading authority on video surveillance, delved deeper into this topic. The article states that while 'fever detection' is new to most people, the technology has been practiced for many years and has a series of international standards. These standards have been published by the world’s top two international standards groups.
The International Electrotechnical Commission (IEC) and The International Standards Organization (ISO) are both based in Geneva, Switzerland. The IEC is the world’s top standards organization for all electrical, electronic, and related technologies. The ISO publishes international standards for a broader range of products.
According to the IEC, the best area to take a body temperature reading is the inner eye. That’s due to its placement over an important artery. They go on to say that other body areas are unreliable.
“Facial thermography of surface areas other than the region medially adjacent to the inner canthi is unreliable, and may be complicated by perspiration, facial skin flushed from exertion, etc.”
So, an elevated body temperature does not always mean someone is sick. The person could have been sitting in a hot car. They could have been exercising.
Their findings also state that hair, glasses, and hats can obstruct the face and therefore cause inaccurate readings. Face masks can also be an obstruction because they can lead to warmer than usual readings due to warm breath exhalations. However, many firms who are marketing thermal-imaging cameras as effective tools for detecting fevers fail to disclose that glasses and masks can impact accuracy or the importance of reading the inner eye.
The IEC states to achieve the most accurate readings, faces must be positioned parallel to the camera. “Cameras must be parallel in order to maximize the number of pixels in the face image, which should be a minimum 240 by 180.”
The article elaborates on this. “If a face on camera needs to be 240 pixels wide and a real face is about six inches wide, effectively a 400-pixel camera should have a field of view no wider than 10 inches. However, some cameras are only getting around 50 pixels per face at best, a fifth of the IEC recommendation.”
One of the most important factors to consider is how and where the cameras are being used. In the article, “The ISO/TR 13154:2017 and IEC 80601-2-59:2017 standards specifically state that fever screening is deployed under indoor conditions. However, this hasn’t stopped some companies from claiming to have outdoor fever detection solutions.”
Lighting and ventilation are also crucial, as the article states, “The IEC says the type of lighting matters, as it could affect accuracy. Specifically, bright lights that emit heat, such as incandescent or halogen, should be avoided.”
Air conditioning vents, drafty windows, and radiant heaters can also cause cooling or heating effects of the face and can impact accurate readings, as well. Ideally, the temperature should be between 20°C to 24°C (68°F and 75°F) and humidity from 10% to 50%.
The ISO and IEC list several recommendations for the screening process that could affect results. They explain that screening areas should not be placed near restrooms for two reasons:
In addition, they suggest not using a dark or reflective backdrop behind the person being screened. Finally, people should be screened one-at-a-time to get the best possible shot of the inner eye.
IPVM wanted to put the cameras to the test. They purchased Dahua’s thermal temperature monitoring system and conducted several experiments. They found that Dahua's normal temperature readings do a great job of giving people peace of mind, but their testing showed that it comes with a significant risk of missing real fevers. In fact, their tests revealed the system regularly missed elevated skin temperatures of 100-102 degrees, especially when the subjects were moving or wearing hats.
The article states, “To simulate fever, our subjects elevated the temperature of the forehead using a hot water bottle in the low 100s° F. The subject was then measured by the Dahua camera, and finally by a handheld IR thermometer (Extech IR200).”
In one test, the Dahua camera measured the subject at 99.6°F, but the thermometer read 100.6°F, a full degree difference. The same results occurred on multiple individuals.
When a subject put on a hat and sunglasses, the Dahua system measured him at 100°F, but he registered 102°F using the IR thermometer. Similar results occurred when subjects wore a knit beanie, ski mask or balaclava, as well as when hair was covering the forehead.
Additional testing was conducted on moving subjects, with very similar results.
“During testing, subjects were frequently measured significantly lower when moving than standing, with temperatures increasing slowly the longer subjects remained in the scene. For example, when walking normally, one subject was measured at 97.6°F. However, when remaining still for 3-4 seconds, his temperature is measured at 98.2°F.”
The testing also revealed something concerning. Dahua uses what they call a “compensating algorithm.”
“Moreover, our testing revealed that the remarkably normal temperature readings Dahua produces are artificially enhanced by what Dahua described to us as a ‘compensating algorithm’ that we found dynamically increases the offset to transform low / bad reads into more normal temperature readings.”
Essentially, the lower the temperature reading the camera has on a face, the more they increase the offset to make the reading look closer to “normal.”
IPVM also conducted tests on Sunell's Panda Cam. On the plus side, "Our testing of Sunell's body temperature measurement camera showed people were measured within ±0.3°F of an IR thermometer in most normal cases, plus Sunell monitors over a much wider range, and, unlike Dahua, did not as aggressively try to turn bad reads into normal ones."
However, there are some negatives. "Sunell only attempts to measure forehead temperatures which (1) increases problems when the forehead is obscured (e.g., hats and hair) and (2) sacrifices the ability to measure the more typically accurate inner canthus / eye area. Secondly, human temperature offsets required regular readjustment, with readings skewing lower as days became warmer."
For the IPVM test, a hot water bottle was placed on the subject's cheek and heated to over 102°F. Because the camera only measures the forehead, the subject measured 98°F.
Sunell explains the reason for the forehead detection rather than measuring the inner eye. "It allows them to focus on larger numbers of subjects using their wider field of view than other cameras, as it requires fewer pixels on target than measuring the inner canthus." The result? Inaccurate measurements.
Other news organizations and experts in the digital world have also weighed in on the topic. Here’s a summary of their findings.
Not all thermal cameras are created equal. Even the camera manufacturers warn their products should only be used as first step in the screening process. Accuracy is key issue.
The Electronic Frontier Foundation (EFF) is the leading nonprofit organization defending civil liberties in the digital world. In their article “Thermal Imaging Cameras are Still Dangerous Dragnet Surveillance Cameras,” they do agree there is a need to monitor a person’s body temperature under certain circumstances. However, thermal-imaging cameras are not the way to do it.
“The cameras typically only have an accuracy of +/- 2 degrees Celsius (approximately +/- 4 degrees Fahrenheit) at best. This is cause for major concern. With such a wide range of variance, a camera might read a person's temperature as a very high 102.2 degrees Fahrenheit when they are actually running an average 98.5 degrees Fahrenheit. What’s more, human temperatures tend to vary widely, as much as 2 degrees Fahrenheit. Not only does this technology present privacy problems, but the problem of false positives cannot be ignored.”
Another factor to consider: a large number of people who are spreading the coronavirus are asymptomatic or have mild symptoms that would avoid triggering a thermal-imaging camera, even if it were 100% accurate at detecting fevers. A recent study in Iceland looked at tests from a large portion of the population. It found that 50 percent of everyone who tested positive were asymptomatic.
A Reuters article states, “Thermal cameras do not measure absolute temperature but rather the difference in energy emitted between one object and another. The systems require regular re-calibration, for example, to handle a factory shift that begins on a cold morning when workers are filing in from outdoors versus an afternoon shift when the sun has warmed the surroundings.”
From the Washington Post, “Such cameras have been deployed at borders and airports in Asia since the SARS epidemic — and public health authorities have previously warned that temperature screenings are no silver bullet.”
The article continues, “The World Health Organization says on its website that ‘temperature screening alone may not be very effective as it may miss travelers incubating the disease or travelers concealing fever during travel, or it may yield false positive (fever of a different cause).’
Finally, from NBC News, “The problems with this technology, according to thermal imaging and virus surveillance experts, is that thermal imaging is an imprecise method for scanning crowds and doesn't measure inner-body temperature.”
As you can see, there are many factors that can impede the accuracy of using a thermal-imaging camera to detect fever. Environmental conditions, facial obstructions, and other medical conditions can all elicit false readings. Even the technology itself is an issue. So, companies that are interested in using them as a solution should do so with caution.