How Does an IR Camera Work: Microbolometer Sensor Guide
An IR camera turns invisible infrared light into an image humans can read. The physics of how that happens differs sharply between the two main IR camera types. This guide focuses on the detector technology, the signal processing chain, and the conversion from raw sensor output to the colored thermal image on screen. The detector array is the heart of the camera and the single component that defines what the device can see, how accurately it can measure, and what it costs. Understanding the detector helps buyers and operators read spec sheets and make sense of why some cameras cost ten times what others do.
The detector at the heart of a thermal camera
Most consumer and inspector-grade IR cameras use an uncooled microbolometer array. Each pixel is a tiny resistor suspended over a substrate by thin support arms. Infrared photons hit the pixel, heat it slightly, and change its electrical resistance. The camera reads voltage across each pixel and converts the readings to temperature values.
Microbolometer materials and construction
Microbolometers use vanadium oxide or amorphous silicon as the temperature-sensitive material. The pixels sit on micro-supports that thermally isolate them from the substrate so that small heat inputs produce measurable resistance changes. The whole array sits in a vacuum package because air conduction would equalize the pixels too quickly and ruin the measurement.
How a single pixel responds to infrared light
An incoming infrared photon transfers energy to the pixel material. The pixel temperature rises by a tiny fraction of a degree. The resistance changes proportionally. The readout circuit measures the resistance change and assigns a numerical value to that pixel. The whole sequence happens hundreds of times per second across hundreds of thousands of pixels.
Spatial resolution and pixel pitch
Modern microbolometer arrays use pixel pitches of 17 microns or smaller. Smaller pixels mean more pixels can fit in the same sensor area, which raises resolution. A 320 by 240 sensor with 17-micron pixel pitch fits in a 5.44 by 4.08 millimeter active area. A 640 by 480 sensor at the same pitch needs 10.88 by 8.16 millimeters of active area, which drives up sensor cost.
Why thermal cameras use germanium lenses
Ordinary glass blocks long-wave infrared light. Thermal cameras use germanium, chalcogenide glass, or similar materials that transmit 8 to 14 micron radiation. The lenses are expensive because the materials and the polishing tolerances differ from visible-light optics. The optics also need anti-reflection coatings tuned to the IR band, which is a different process from visible coatings.
Cooled versus uncooled detectors
Scientific thermal cameras sometimes use cooled detectors that operate at cryogenic temperatures. Photon detectors made of materials like indium antimonide or mercury cadmium telluride produce sharper images and lower noise than uncooled microbolometers but require a Stirling cycle cooler or liquid nitrogen. The cost reaches tens of thousands of dollars per unit, and the cooler adds size, weight, and maintenance.
Reading the array: rolling versus snapshot
The readout circuit reads each pixel in turn during a frame cycle. Most microbolometer arrays use a row-by-row rolling readout. Each frame thus represents a slightly different moment in time across rows, but at typical frame rates of 9 Hz to 30 Hz the effect is negligible for static building scans. Faster moving targets reveal motion artifacts that high-speed thermal cameras avoid by using snapshot readouts.
Non-uniformity correction
Microbolometer pixels have slight manufacturing variations that would otherwise produce a noisy image. Cameras run a non-uniformity correction during operation, typically using a shutter that briefly closes in front of the sensor to capture a uniform reference. The camera then adjusts each pixel’s output to match the reference, which produces a clean image. Most users notice this as a brief click and screen freeze every minute or two.
Converting voltage to temperature
The raw output of the sensor is voltage. Converting voltage to absolute temperature requires calibration against known temperature targets across the operating range. Factory calibration data is stored in the camera and used to interpolate temperature for any measured voltage. Calibration drift over time is why pro cameras get sent in for annual recalibration to maintain their accuracy specification.
Emissivity correction in the math
The camera measures total infrared radiation arriving at the sensor, which includes both emitted radiation from the target and reflected ambient radiation. To extract the target’s true temperature, the camera applies an emissivity correction. The user sets an emissivity value, typically 0.95 for building materials, and the camera does the math. Polished metal at emissivity 0.05 requires either a very different setting or surface preparation with paint or tape.
Atmospheric absorption corrections
Long-wave infrared light passes through clear air fairly well but is absorbed by water vapor and carbon dioxide. For distances under 10 meters, the absorption is negligible. For longer distances, pro cameras apply a correction based on humidity, atmospheric temperature, and target distance. Operators enter these values in the camera menu before scanning at distance.
NETD: what thermal sensitivity means
Noise equivalent temperature difference, or NETD, expresses the smallest temperature difference the camera can reliably distinguish. Consumer cameras quote 50 to 100 millikelvins. Pro cameras quote 30 to 50 millikelvins. Lower NETD means the camera can see subtler patterns. The number reflects sensor noise floor and is one of the biggest spec-sheet drivers of price.
Frame rate and the 9 Hz export cap
Uncooled thermal cameras sold to U.S. consumers are capped at 9 Hz refresh under ITAR rules. The cap controls the rate of image updates and limits the camera’s usefulness for fast-moving targets. Cameras with higher refresh rates exist but require export licensing. For static scans of buildings and electrical panels, 9 Hz is fine and most users do not notice the limit.
How the image gets to the screen
After the array readout and calibration math, the camera applies a color palette to the temperature data. Iron, rainbow, grayscale, and arctic are common palettes. The result appears on the camera screen, often with a temperature scale on the side and a center-spot temperature readout. The colors are a visualization choice that does not change the underlying data. Operators can switch palettes after the fact in radiometric image files.
Image fusion with a visible camera
Pro thermal cameras pair the thermal sensor with a visible-light sensor mounted alongside. The camera registers the two images by software and lets the operator blend them at adjustable transparency. Some manufacturers overlay just the edges of the visible image on the thermal image, which keeps the thermal data dominant while still providing context. The technique helps the operator identify what they are looking at.
Storage and radiometric image files
Pro cameras save radiometric image files that preserve temperature data for every pixel. Reporting software loads these files and lets the operator probe individual pixels, adjust the temperature span, change palette, and modify emissivity after the fact. Consumer cameras often save flat JPEGs that capture the visual representation but lose the underlying temperature data. The format choice limits what the operator can do later.
Active near-infrared cameras work differently
Active NIR cameras used in security work do not use microbolometers. They use ordinary CMOS image sensors with the IR-cut filter removed and IR LEDs added for illumination. The sensor responds to near-infrared wavelengths around 0.7 to 1.0 microns. The LEDs flood the scene with light invisible to humans, and the sensor records the reflected light. The result is a grayscale video feed. The technology differs completely from thermal imaging despite sharing the “infrared camera” label. Our coverage of night-vision infrared cameras covers the contrast in detail.
Calibration accuracy expectations
Consumer cameras typically quote accuracy of plus or minus 2 degrees Celsius or 2 percent. Pro cameras quote plus or minus 1 degree or 1 percent. Scientific cameras can quote tighter still. Accuracy depends on the calibration process at the factory, drift over time, and operator settings like emissivity. For anomaly detection, accuracy rarely limits the work. For absolute temperature measurement, it matters.
What buyers should look for on a spec sheet
Sensor type and resolution define what the camera can see. NETD defines how subtle the patterns it can detect. Spectral range should be 8 to 14 microns for a true thermal camera. Accuracy and operating temperature range matter for outdoor work. Frame rate at 9 Hz is the U.S. consumer cap. Storage format and radiometric support matter for reporting. The home inspection tools pillar covers how these specs translate to actual inspection workflows.
The wavelength bands that matter
Thermal cameras work in the long-wave infrared band of 8 to 14 microns because room-temperature objects radiate most strongly there, by Wien’s displacement law. Hotter objects shift their peak emission to shorter wavelengths, which is why some industrial cameras use mid-wave infrared bands at 3 to 5 microns for high-temperature targets. Near-infrared at 0.7 to 1.0 microns is reflected-light territory for security and trail cameras, not heat sensing.
Why room-temperature objects show up clearly
An object at 20 degrees Celsius (293 Kelvin) emits peak radiation around 10 microns. That sits right in the middle of the 8 to 14 micron LWIR band used by thermal cameras. The matching of emission peak to detector band is why LWIR cameras work so well for everyday building inspection, electrical scanning, and human body detection. The same camera would not be optimal for measuring high-temperature furnaces, which radiate most strongly at shorter wavelengths.
Shutter calibration in operation
Most uncooled thermal cameras include a mechanical shutter that closes briefly during operation. When closed, the shutter presents a uniform temperature surface to every pixel. The camera reads each pixel’s response to that uniform input and adjusts the pixel-level corrections. The process takes a fraction of a second and runs automatically every minute or two. Users hear a faint click and see the live image briefly freeze.
Signal processing pipeline summary
The pipeline reads voltage from each pixel, applies non-uniformity correction, applies factory calibration to convert voltage to temperature, applies emissivity and ambient corrections, formats the data into a temperature array, and finally applies the user-selected palette for display. Each step adds value and matters for image quality. The same hardware can produce radically different results depending on calibration quality and signal processing tuning.
How firmware updates change camera behavior
Manufacturers issue firmware updates that improve image processing, add new palettes, refine calibration math, and sometimes raise accuracy specifications without hardware changes. Updates ship through manufacturer software for pro cameras and through app stores for phone-attached units. Buyers should check firmware status periodically and apply updates that improve their workflow.
References
- Professional Home Energy Assessments — U.S. Department of Energy
- Certified Infrared Inspector Program — InterNACHI
- Electrical Infrared Inspections — National Fire Protection Association
- ASHI Standards of Practice — American Society of Home Inspectors