How Thermal Cameras Work: Physics, Sensors, and Optics
The technical question of how thermal cameras work has a satisfying answer once a reader can hold four ideas in mind at the same time. The first is Planck’s radiation law, which says every object above absolute zero radiates electromagnetic energy in a spectrum determined by its temperature. The second is the Stefan-Boltzmann scaling, which says the total radiated power rises with the fourth power of temperature. The third is the microbolometer focal plane array, which is the actual physical sensor inside the camera. The fourth is the germanium optical path, which is required because ordinary glass is opaque to the wavelengths the sensor needs to see. Put those four together and a thermal image stops being mysterious. This guide treats each piece in technical depth aimed at readers who want more than the plain-language explanation.
Planck’s Law and the Long-Wave Infrared Band
Max Planck’s 1900 derivation of blackbody radiation laid the foundation. Any object at temperature T radiates a continuous spectrum of electromagnetic energy, and the peak wavelength of that spectrum is determined by T through Wien’s displacement law. For an object at human-comfortable room temperature near 295 Kelvin, the peak emission wavelength sits around 9.8 micrometers, squarely in what optical engineers call the long-wave infrared band — roughly 7 to 14 micrometers. Building materials, human skin, furniture, and most everyday surfaces around 270 to 320 Kelvin all radiate most strongly inside this 7-to-14-micrometer window.
That alignment is why residential and industrial thermography cameras are designed specifically for long-wave infrared. A camera optimized for the medium-wave band (3 to 5 micrometers) sees better in hotter applications — gas-turbine surveys, industrial process monitoring — but loses sensitivity at building-temperature ranges. A camera in the near-infrared band (0.7 to 1.4 micrometers) sees nothing useful at room temperature because almost no thermal radiation is being emitted in that range. Long-wave infrared is the right band because Planck’s law puts the peak emission of building materials inside it.
Stefan-Boltzmann and Why Small Temperature Differences Are Visible
The total radiated power from a surface scales as the fourth power of absolute temperature. At room temperature, that fourth-power scaling translates to roughly 1.3 percent change in emitted radiation per Kelvin of temperature change. A 1 K difference between an insulated wall section and an uninsulated one at 295 K produces a measurable difference in the long-wave infrared flux arriving at the sensor — small in absolute terms but well above the noise floor of a modern microbolometer array. The sensor’s noise-equivalent temperature difference (NETD) specification is the relevant figure: residential-grade cameras typically achieve 50 to 100 millikelvin NETD, meaning they can resolve temperature differences of half a degree Fahrenheit or better. Trade-association guidance from InterNACHI on infrared thermography references NETD as the key sensitivity figure inspectors should evaluate when selecting a residential camera.
The pillar guide on the tools used during professional residential inspections covers how this sensitivity drives the diagnostic value of a thermal scan during a home inspection.
The Microbolometer Focal Plane Array
The sensor itself is a focal plane array of microbolometers — thousands of microscopic resistors made from vanadium oxide or amorphous silicon, suspended on tiny silicon nitride bridges over a CMOS readout substrate. Each microbolometer absorbs incident long-wave infrared photons, warming by a tiny fraction of a degree. The resistance of the vanadium oxide or amorphous silicon changes measurably with that temperature, and the underlying CMOS circuit reads the resistance change as a voltage. The whole array is housed inside a vacuum-packed cavity to eliminate convective heat transfer between the microbolometer and its surroundings, which would otherwise destroy the signal.
Typical resolutions for residential-grade cameras run 160 by 120 pixels at the entry level, 320 by 240 in the mid-range, and 640 by 480 or higher in professional and research-grade equipment. Each pixel produces one temperature reading per frame, and the camera typically refreshes at 9, 30, or 60 hertz. Higher frame rates require more sensor cooling and faster readout electronics, which drive the price up sharply. The 9-hertz limit on consumer-grade cameras is partly a U.S. Export Administration Regulation classification boundary — cameras above 9 hertz fall under tighter export controls because of potential military applications.
Why Germanium Lenses
Ordinary optical glass is highly opaque in the long-wave infrared band. A silicon dioxide lens that transmits 90 percent of visible light transmits less than 1 percent of 10-micrometer infrared. Germanium, an element on the same column of the periodic table as silicon, is one of the few materials that transmits efficiently across the entire 7-to-14-micrometer window. Modern thermal-camera lenses are machined from germanium, sometimes coated with diamond-like carbon for scratch resistance, and the resulting optic looks slightly orange or gray rather than clear. The cost of germanium is one of the larger contributors to the overall price of a thermal imager — a 640-by-480 camera with a high-quality germanium lens system costs several thousand dollars partly because the lens itself runs several hundred.
Some cheaper smartphone-attached thermal cameras use chalcogenide glass instead of pure germanium, achieving lower transmission but adequate performance for low-resolution sensors. These are the cameras priced below 400 dollars that connect to a phone’s USB-C port. Their sensor resolution and lens quality both limit diagnostic depth, but for homeowner orientation and obvious-finding hunting they work well enough.
Emissivity Corrections
The signal arriving at the sensor depends not just on the temperature of the target surface but also on that surface’s emissivity. A perfect blackbody emits 100 percent of the theoretical Planck spectrum at its temperature; a polished aluminum surface emits about 5 percent of it. Building materials — drywall, wood, brick, painted surfaces — typically run between 0.85 and 0.95 emissivity, which is high enough that the camera’s default settings produce useful temperature readings without correction. Polished metals, glass at oblique angles, and reflective surfaces all require either an emissivity correction in software or a piece of high-emissivity tape stuck to the target before reading.
Inspectors documenting findings on standard building surfaces rarely adjust emissivity in the field. Inspectors documenting electrical panel interiors with polished copper bus bars often do. The error from uncorrected emissivity on copper can run 20 degrees Fahrenheit or more, which is enough to either miss a finding or invent one. The pillar article on what thermal imaging actually means as a technology category covers the emissivity issue from a different angle.
Atmospheric Attenuation Over Distance
Long-wave infrared transmits well through dry air over short distances. Over distances longer than several meters, water vapor in the air starts to absorb specific wavelengths inside the 7-to-14 band, and the camera reads slightly cooler than the actual target. For residential scans inside a building, this attenuation is negligible. For outdoor scans at distance — a thermal survey of a building exterior from across a parking lot, for instance — the attenuation becomes significant enough that professional thermographers compensate in software using measured ambient humidity and target distance. Department of Energy guidance on building thermal scans mentions distance compensation for any exterior survey beyond about 10 meters.
Why the Image Refresh Looks Smooth
Microbolometers have a thermal time constant of roughly 10 milliseconds — the time it takes the sensor element to settle to a new temperature after the incident radiation changes. At a 30-hertz frame rate (33 milliseconds per frame), the sensor settles between frames and the image looks smooth to the eye. Lower-cost 9-hertz cameras can produce visible lag during fast panning because each frame integrates across a longer window. This is the practical reason inspectors prefer 30-hertz cameras for live walk-through use, even though 9-hertz cameras are perfectly adequate for static documentation.
Non-Uniformity Correction and Internal Shutter
Individual microbolometer elements in the focal plane array do not respond uniformly to incident radiation. Slight manufacturing variations, age-related drift, and ambient temperature changes all introduce per-pixel offset that would otherwise show as fixed-pattern noise in the image. Modern thermal cameras include an internal mechanical shutter that periodically (every 30 to 90 seconds during operation) closes over the sensor for about 100 milliseconds, presenting a uniform-temperature reference scene. The camera measures each pixel’s response to the reference and stores a correction offset that is then applied to subsequent frames.
This routine, called non-uniformity correction or NUC, produces the brief shutter-click sound and the momentary screen freeze that users notice during operation. The NUC process is automatic and cannot be disabled in most consumer and entry-professional cameras. Higher-end research cameras allow manual NUC control because some scientific applications require exact synchronization of frames with external events that the automatic NUC schedule would interrupt.
Radiometric Storage and Per-Pixel Temperature Data
A radiometric thermal camera stores per-pixel temperature values alongside the displayed false-color image. The format is typically a proprietary file (FLIR’s .seq and .csq, Fluke’s .is2, and others) that contains both the visualization and the underlying radiometric data. Post-processing software can extract specific temperature readings from any pixel in a saved image, apply different emissivity corrections, adjust span and level for clearer visualization, and generate quantitative analysis reports from the captured data.
Non-radiometric cameras store only the displayed image as a standard JPEG or PNG file with the false-color visualization baked in. The per-pixel temperature data is lost at capture time. Most smartphone-attached thermal cameras and the cheapest entry-professional models are non-radiometric. Inspectors generating formal documentation reports require radiometric capture because subsequent report writing often involves measuring temperatures from points that were not deliberately captured live during the original scan.
Lens Focal Length and Working Distance Trade-Offs
The physical focal length of the germanium lens determines the camera’s field of view and instantaneous field of view. A wider-angle lens (shorter focal length) captures more scene per frame but produces lower spatial detail per pixel. A telephoto lens (longer focal length) captures less scene but resolves smaller features at distance. Residential inspection cameras typically use a 19-degree to 25-degree wide-angle lens because the working distances inside a building are short and the inspector benefits from seeing larger sections per frame.
Some higher-end cameras include interchangeable lenses for specialty applications. A telephoto lens makes sense for exterior building surveys from across a parking lot or for industrial substation scans where the inspector must keep safe working distance from energized equipment. A macro lens makes sense for printed-circuit-board diagnostics. Residential inspection rarely needs lenses other than the wide-angle that ships standard with most handheld inspection cameras.
Software Algorithms That Enhance the Raw Sensor Data
Modern thermal cameras apply several software algorithms to the raw sensor data before displaying or storing the image. Image fusion blends the thermal data with a visible-light photograph taken simultaneously through a parallel optical path, producing an output where the thermal information overlays visible-light edge detail. Super-resolution interpolation increases the apparent pixel count of the displayed image by combining multiple frames or applying machine-learning enhancement. Noise filtering reduces frame-to-frame variation that comes from electronic noise without significantly affecting actual thermal information. Each algorithm trades some fidelity for some readability, and the choices reflect manufacturer assumptions about how the camera will be used.
For diagnostic reporting, the cleanest data comes from minimally processed images. Cameras targeted at professional inspection use typically offer a “raw” or “minimally processed” mode that preserves sensor data without aggressive enhancement. Cameras targeted at consumer use typically default to heavy enhancement that makes the live preview look better but may obscure subtle thermal differences that matter diagnostically. Working inspectors learn the specific processing pipeline of their chosen camera and configure it to support the diagnostic interpretation they apply during report writing.
References
- Department of Energy reference on infrared thermography physics for building scans — U.S. Department of Energy
- InterNACHI Standard of Practice and technical notes on infrared thermography — InterNACHI
- ASHRAE technical standards on building thermal diagnostics — ASHRAE
- EPA moisture-control reference for surface emissivity in building diagnostics — Environmental Protection Agency