There are many applications in electronics that promise to make our lives more comfortable and safer. Thermal imaging cameras are increasingly seen to hold such promise, particularly since the onset of the Covid-19 pandemic.

The rapid global spread of the virus presented an unprecedented challenge to public health, food systems, and the world of work and led to a dramatic loss of human life worldwide. It decimated jobs and placed millions of livelihoods at risk. In the face of this crisis, finding a way to measure body temperature quickly from a distance — including the ability to scan groups of people simultaneously, without the need to interfere with their activity — becomes very important for managing the pandemic. This is because an elevated temperature is one of the common symptoms of viral infection, and therefore, the ability to efficiently identify people who are running a fever can help to confine the spread of the virus that causes Covid-19.

Thermal imaging sensors are at the core of the technology that gives us this ability.

But what are thermal imaging sensors, and how do thermal imaging cameras let us measure core body temperature from a distance? Thermal imaging sensors are microelectromechanical system (MEMS) chips that include an array of detectors sensitive to impinging long-wave infrared electromagnetic radiation (LWIR), in the range of 8- to 14-µm wavelengths. All objects above 0 Kelvin and all living organisms radiate in this spectral range, and the intensity of this radiation is representative of their surface temperature.

Unlike visual light of wavelengths between 400 and 700 nm, LWIR radiation is invisible to the human eye. However, the detectors forming the array of a thermal imager are able to respond to the incident infrared heat by changing their properties in a measurable way, e.g., by changing their resistance or by generating a voltage output due to the Seebeck effect. These changes are amplified and digitized by a readout circuit, and the resulting digital codes are eventually transformed into temperature values.

Therefore, each thermal detector measures the surface temperature of the objects in its own instantaneous field of view — the fraction of the world from which it can perceive emitted heat radiation. The ordered two-dimensional array of detectors then yields a two-dimensional array of pixels — akin to a gray-scale visual image, but in this case, each pixel represents how hot that part of the scene is, instead of how bright it is.

Beyond a simple analogy, the reference to a gray-scale image is useful because one can easily intuit that a thermal image can also be processed and analyzed by more or less standard techniques of visual image processing. It can also be displayed as an informative heat map, whereby for each temperature value, there is a specific associated color.

An example of a visual image, gray-scale thermal image, and a colored thermal image — all three pertaining to the same scene — is shown in Figure 1.

Figure 1: Visual camera image versus thermal camera image

These thermal images are captured by a special camera module using the LWIR thermal imaging chip shown in Figure 2.

Figure 2: Meridian Innovation 80 × 62 resolution LWIR thermal imaging chips and module

We stated earlier that thermal imagers measure the surface temperature of the objects or subjects in the field of view. The question now is whether, and how, we can use the information in a thermal image to determine the core body temperature of a human subject.

Before answering that, let us recall that traditionally, core body temperature is measured by a mercury or digital thermometer (Figure 3) that must be in physical contact with a concealed area of the body that has a temperature very close to or the same as that of the internal organs. Typical places of contact are inside the rectum (rectal thermometers, which are the most accurate of the contact thermometers), under the tongue (oral thermometers, about 0.3°C lower than the rectal temperature readout), or the armpit (axillary thermometers, about 0.6°C lower than the rectal temperature readout).

The practical limitations of the contact thermometers have been overcome by the development of different types of non-contact, infrared-sensing thermometers, which can be distinguished by their form factor and the way they must be positioned to perform the measurement — for example, tympanic thermometers, positioned in the distal ear canal; temporal artery thermometers, slid across the superficial temporal artery on the side of the forehead; or forehead thermometers (Figure 4), placed right in front of the forehead.

Figure 3: Boots contact thermometer (L) (Source: Meridian Innovation)

Figure 4: Braun forehead thermometer (R) (Source: Meridian Innovation)

There are three common factors among these thermometers:

A thermal imaging camera, on the other hand, is built around a multipixel thermal imaging sensor as described earlier. The design and manufacturing process used in the formation of the detectors not only allows miniaturization, which favors packing more detectors per die (e.g., 5,000 pixels or more), but also results in more sensitive detectors with a short response time, which allows one to capture not just static scenes but thermal video streams at frame rates anywhere from 1 to 30 frames per second.

Together with a specialized lens that typically has an angular field of view of at least 40°, the technology allows one to capture complex, dynamic thermal scenes, including multiple people at the same time, at distances ranging between 50 cm and several meters.

All this seems to make it possible to quickly measure body temperature for groups of individuals at the same time, from a distance, without stopping them — but is it really enough? Clearly not.

For starters, the detectors can sense only the temperature on the surface of the skin. Similarly to the forehead thermometer, the core body temperature must be calculated based on some biophysical model of the human body. The question that now arises is which pixel to take as representative of the skin temperature of interest. This is a non-trivial matter because the thermal imager resolves the face of a human in multiple pixels, with temperature readouts that can vary by more than 1°C.

Additionally, factors such as imperfections in the lens; the viewing angle between the camera and the subject; and the presence of a face mask, a face-occluding hair style, glasses, etc., all contribute to some inaccuracy in the temperature readout and make its interpretation a lot more complex than in the case of a simple forehead or in-ear IR thermometer.

Here is where the crucial role of thermal image processing and analytics comes into play. The various inaccuracies are mostly of systematic character and can be compensated for, taking into account the “excess” information available in the thermal image. This, however, makes the design of a thermal imaging camera for fever detection a rather complex endeavor, requiring a detailed understanding of the thermal imaging sensor properties, a good grasp of thermography and some biophysics, and, of course, a proficiency in image and data analytics.

Typical systems for temperature scanning include both visual and thermal imaging sensors. The visual image stream is analyzed for face detection and face-mask detection, and the resulting regions of interest are mapped to the thermal data stream for temperature analysis.

The result of such processing is shown in Figure 5, where three people are identified in a given pair of visual and thermal frames, and each person’s estimated core body temperature is annotated on the visual frame. However, scanning for fever becomes significantly more difficult when the privacy of the individuals must be preserved and a visual image sensor therefore cannot be used on the system. In that case, the entire process of subject detection and subsequent core body temperature estimation must be based solely on the thermal video stream.

Figure 5: Meridian Innovation multiple-people fever-detection camera solution (Source: Meridian Innovation)

More details on the principle of how it works can be found in the “Research Note on Fever Detection,” by Dr. Stanislav Markov of Meridian Innovation, at erp.meridianinno.com/documents/On_Fever_Detection_english.pdf.

The imperative for a timely delivery of a thermal imaging camera for the detection of elevated temperature and the complexity of building a good solution, as described above, have led to joint efforts to provide a number of reference designs for OEMs and ODMs. An example of such a reference design is provided by Arrow Electronics and is presented in greater detail at www.arrowopenlab.com/HkOpenLab/solutions/2021E001.pdf.

At the reference design’s core is the novel thermal imager by Meridian Innovation (meridianinno.com). The imager is mass-produced via a CMOS/MEMS technology and is a hybrid between conventional microbolometers and bulk thermopile in terms of pixel design. Other than the thermal imager, the reference design includes an ambient sensor and a time-of-flight sensor to estimate the distance to subjects. It benefits from AI-enhanced application software, which improves the accuracy of the thermal readout and core body temperature calculation, as well as the quality of the thermal image for visualization. The choice of cost-effective components together with the software development kit, which helps with the most significant hurdles — including the calculation of core body temperature — lowers the barrier for adoption of the design and helps to meet the demands for thermal imaging cameras that can detect a fever.

Thermal imaging sensors are now readily available and affordable for mass-market products. Accessible reference hardware designs and software solutions facilitate adoption of the technology and enable a fast development cycle for thermal imaging cameras, including cameras for automatic fever detection. Such products can help manage the pandemic and present a compelling story in which innovative technology fosters safer and better living.

You must Register or Login to post a comment.

This site uses Akismet to reduce spam. Learn how your comment data is processed.