While thermal infrared imaging technology was once limited to defense and government-funded applications, it has become more mainstream over the past few decades. The growing awareness of the technology’s industrial benefits has primarily been driven by the increasing popularity of hand-held thermal cameras for maintenance and electrical troubleshooting. Then came the COVID-19 crisis last year, and with it an onslaught of thermal imaging solutions flooded the market. A variety of thermal imaging sensor technologies and cameras have been used for fever detection, although, technically speaking, the sensors only measure surface skin temperature. Commonly opaque packaging plastics are often more transmissive in the SWIR than in the visible range, allowing SWIR cameras to easily detect the fill level of contents that are less transmissive or even strongly absorptive of SWIR wavelengths (left). False color palettes are used to highlight or accentuate temperature transitions for the human observer, without altering the temperature values of the underlying pixel (right). Courtesy of MoviTHERM. Nonetheless, infrared and thermal imaging is still mysterious to many end users. Even skilled machine vision integrators may struggle with the implementation of nonvisible-imaging technology. This is not surprising, since humans lack the ability to visually perceive temperature. To better comprehend what infrared and thermal cameras can do, users must understand how the cameras work and the physics involved. In contrast to standard machine vision cameras that operate mainly in the visible spectrum between 400 and 700 nm, infrared and thermal camera technology covers a significantly wider spectral range subdivided into three regions: the shortwave infrared (SWIR) between 0.9 and 1.7 µm, the midwave infrared (MWIR) between 3 and 5 µm, and the longwave infrared (LWIR) between 8 and 14 µm. The spectral band specification is defined by the characteristics of the detector technologies used for the various cameras. The spectral bands are derived from the wavelengths to which the detector materials are sensitive. Physics literature may segregate these infrared spectral bands differently based on scientific principles. Into the SWIR Many common applications can benefit from detection in each waveband. Not all involve temperature measurements. Some exploit the physics of materials science with respect to spectrally selective reflection, absorption, and/or transmission. Commonly opaque packaging plastics, for example, are more transmissive in the SWIR than in the visible range, allowing SWIR cameras to easily detect the fill level of contents that are less transmissive or even strongly absorptive of SWIR wavelengths. The result is a SWIR image with a contrast that is good enough to permit inspection. SWIR technology also applies to agriculture, where it can monitor crop and plant health, detect bruising, or measure the sugar content of fruit. All of these applications use some form of spectral reflectivity, absorption, or transmission as the basis for the underlying inspection method. When measuring temperatures with a SWIR camera, it is important to understand that in the SWIR spectral region most of the signal is still generated from reflected light and not from radiated infrared energy. This can be illustrated by a standard machine vision application that relies on visible daylight. Measuring temperature with a SWIR camera takes a considerable amount of thermal energy to overcome the reflected light and register as radiated energy at the sensor. So, performing a temperature calibration on a camera with a SWIR detector usually makes no sense for temperatures below 400 °C. This qualifies SWIR cameras for high-temperature applications such as imaging molten metals or inspecting process welds, among others. Cool under pressure True thermal effects at room temperature or below manifest at wavelengths around 3 µm and above. Imaging devices that are able to capture these effects are more typically considered true thermal cameras. The term “infrared camera” is no longer used to refer to these imagers because most of the signal being captured stems from radiated infrared heat energy. An overview of the electromagnetic spectrum, including applicable infrared wavelengths. Courtesy of MoviTHERM. Scientists and camera makers define the spectral bands of the infrared spectrum differently. The boundaries defined by the latter rely on characteristics of the detector technologies employed for thermal cameras. MCT: mercury cadmium telluride. Courtesy of FLIR Systems Inc. MWIR thermal cameras are well suited for all kinds of thermal imaging applications. However, they have one downside. They are very expensive, with a median price point of around $70,000 for a 640- × 512-pixel detector. These detectors are costly because they must be cryogenically cooled down to about 75 K, or −198.15 °C. The detector material is so sensitive to thermal radiation that the sensor would instantly saturate at room temperature. In modern cameras, cryogenic cooling is accomplished by a closed-circuit Stirling cooler situated inside the camera body. In the past, cooling for these types of cameras was achieved using large gas bottles filled with liquid nitrogen. A more affordable option is a thermal camera equipped with a microbolometer-type detector. Depending on the pixel resolution, detector noise level, and temperature accuracy, these cameras start below $1000 and offer resolutions of 80 × 60 pixels. A microbolometer works entirely differently than a typical photon-capturing detector. Its operation is based on microsize thermoresistive pixels. When these pixels are exposed to infrared radiation (heat), they change their resistance. No cryogenic cooling is required. Some cameras of this type use a thermoelectric cooling element, which is much easier to operate and less costly than cryogenic cooling. Each pixel in an LWIR camera has a physical mass that needs to be warmed up by the captured thermal radiation of the object at which it is pointed. This imposes a fixed time constant that is described by the time it takes for each pixel to properly warm up before the camera reads out the change in resistance. The constant is typically between 8 and 14 ms, depending on pixel size. The downside is that the time constant poses a challenge when it comes to imaging moving objects. Eight milliseconds seems like a short length of time. But depending on the camera’s field of view and the speed of the object that it is imaging, significant motion blur artifacts can occur in the captured image. Motion blur is created when portions of the object pass by detector pixels during the integration time — that is, the time constant. In other words, a pixel may not fully integrate the thermal radiation that it is trying to capture before the object moves over to an adjacent pixel. In turn, this causes a temperature-averaging effect, leading to measurement errors and other issues. Uncooled microbolometer detectors offer a more affordable alternative to cryogenically cooled MWIR cameras. The ability of microbolometers to capture thermal image data is based on microsize thermoresistive pixels that, when exposed to infrared radiation (heat), change their resistance. Courtesy of MoviTHERM. Motion blur is not the only type of blur that occurs in thermal imaging. Since the contrast in a thermal image is caused by changes in temperature, most thermal images appear blurry. This blur is not an effect of focus or the lack thereof. It is a function of physics — thermodynamics, to be more precise. Heat energy flows from warmer regions of higher energy toward cooler regions of lower energy. This behavior is entirely dynamic, and it creates temperature transitions or thermal gradients. Because temperatures in a thermal image are expressed as changes in brightness — white representing hotter regions and black representing colder ones — gray transitions occur between warmer and colder regions. A thermal image of a powered-up electronic circuit. The only circumstance in which such an image appears sharp is when changes in emissivity are present, or when a warmer region is thermally isolated from its surrounding area. Due to this dynamic behavior, which is caused by thermal diffusion, thermal imaging may have more to do with signal processing than with image processing. I/O: input/output. Courtesy of MoviTHERM. These transitions give the appearance of a blurry edge. The effect does not typically occur in standard machine vision applications, which rely more on the effects of light reflected off surfaces or features. This reflective pattern is constant, as is the contrast it creates in an image. The only circumstances in which a thermal image appears sharp is when changes in emissivity are present, or when a warmer region is thermally isolated from its surrounding area. Due to this dynamic behavior, which is caused by thermal diffusion, thermal imaging may have more to do with signal processing than with image processing. Understanding emissivity The property of emissivity may be the single most important phenomenon to understand when dealing with infrared or thermal cameras. Therefore, it is often one of the most talked about subjects in thermography classes and seminars. In short, emissivity describes the ability of a solid to radiate infrared energy. Emissivity is made up of three main components: reflected, transmitted, and radiated energy. The sum of all these components must equal one. Because most materials do not allow transmission of infrared radiation, imaging is mostly concerned with reflected and radiated energy. In this context, deriving a sum of one may make measuring the temperature of a thermally reflective object difficult, if not impossible. For example, trying to discern the temperature of a shiny, stainless steel tank is not considered a viable application for thermal imaging — unless the emissivity of the tank’s surface can be altered. A dull black coating may be applied to a region of the tank to increase its emissivity to 0.9 or higher, if permissible. Through heat conductance, this high-emissivity coating would take on the temperature of the skin of the tank. The coating would then favorably radiate the energy toward the thermal camera, thus allowing for an accurate temperature measurement. Applications that involve low-emissivity surfaces that cannot be altered may need to be measured via a contact method, such as by attaching a physical thermocouple. Another consideration when employing a thermal camera for machine vision is the available spatial resolution of the thermal camera. For commercial applications, the highest resolution is around 1.3 MP, with more affordable cameras offering 640 × 480 or 640 × 512 MP. These capabilities pale in comparison to state-of-the-art machine vision cameras, which can deliver 70- or even 100-MP resolutions. Infrared cameras have some catching up to do. The lens materials used for thermal cameras are exotic. The most typical one is germanium. Standard borosilicate glass blocks mid- and longwave IR light and is therefore not a suitable optical material. Thermal camera manufacturers must calibrate a lens to the very camera into which it is incorporated. Many manufacturers double up as the lens provider for their thermal cameras. So, it is not unusual for them to offer only one to five lens choices per camera, further complicating the design of the imaging system. The situation becomes even more complicated if a thermal camera needs an enclosure to protect it against a harsh environment. In this case, the viewing window must also be equipped with an infrared transmissive glass, such as germanium or another suitable material. Thermal expansion Despite these challenges and drawbacks, thermal cameras are becoming more prominent in both industrial and nonindustrial imaging applications. Several factors have contributed to this growth. The reduction in cost is arguably the biggest contributor. Second is the introduction of a standard communications protocol — GenICAM — as well as a standard physical interface. First came Firewire, and now most cameras are equipped with Gigabit Ethernet. Not long ago, users would have had to use their finest programming skills to implement a proprietary communications interface, with the help of a software development kit, only to learn that the next camera model in the lineup was not compatible. The move by camera manufacturers toward more uniform communications standards has benefited camera sales, as well as resulted in broader adoption by systems integrators and end customers alike. Although thermal camera manufacturers continue to struggle to fully adhere to these standards, the situation has certainly improved. Recent times have also seen the introduction of thermal smart cameras. Despite the world of differences between thermal imaging and standard machine vision, smart cameras will help to further advance thermal imaging’s adoption. Such adoption will only be possible, however, for simple applications that do not need complex image or signal processing. Meet the author Markus Tarin is president and CEO of MoviTHERM. He has extensive experience with visible and nonvisible imaging and has been the lead architect on many development projects for defense, research, medical, and industrial applications.