Selecting an Infrared Thermal Camera Module: NETD, Optics, ISP & SDK Decisions

Industry guidance categorizes NETD roughly as follows: <25 mK = Excellent, <40 mK = Great, <50 mK = Good, <60 mK = Acceptable, <80 mK = Satisfactory. Lower NETD cameras can reveal more subtle thermal differences, rendering more shades of gray (or color) in the image for better contrast. For example, a high-end thermal core with NETD <35 mK will produce a crisper image under the same conditions than a budget core with NETD ~60 mK. In inclement weather (rain, fog) or with small temperature deltas, the low-NETD unit maintains clarity and extends detection range, whereas a higher-NETD unit may show noisy, washed-out imagery.

Beware of spec nuances: Ensure the NETD is specified at the standard scene temperature (around 25–30 °C). Some manufacturers quote NETD at 50 °C to make the number look lower. A reputable spec will list NETD at 25 or 30 °C. For instance, a thermal module might advertise “NETD <50 mK @ 30 °C,” meaning it can distinguish 0.05 °C differences at room temperature. Compare like for like, and when possible, seek real-world image examples at low contrast. Also consider that the lens f-number and detector integration time affect effective NETD – faster lenses (low f/#) and longer integration can improve system NETD.

Typical values: Modern uncooled LWIR microbolometers commonly achieve NETD in the 40–60 mK range for affordable models, and ~20–30 mK for premium models (thanks to advancements like smaller pixel pitches and better readout circuits). For context, FLIR’s compact Lepton module has <50 mK sensitivity, while the latest FLIR Boson+ core boasts <20 mK. Lower than 20 mK is usually only seen in cooled MWIR cameras or cutting-edge uncooled research prototypes. As an OEM, choose a module with NETD that meets the demands of your application – e.g. detecting very slight temperature differences (leaks, insulation faults, faint heat signatures) calls for high sensitivity, whereas detecting larger warm bodies (humans, vehicles) in high-contrast environments might work with moderate sensitivity.

Optics: Field of View and Lens Considerations

The optical lens paired with the thermal sensor determines what the module “sees.” Key optical parameters are the focal length (which sets the Field of View, FOV) and the f-number (speed) of the lens. Thermal camera lenses are typically made of special materials like germanium, zinc selenide, or chalcogenide glass that transmit long-wave IR (8–14 µm) radiation. As an OEM buyer, you may have options to select different lenses for a given core, or request a custom lens to suit your needs.

• Field of View (FOV): This is the angular width and height of the scene captured by the sensor, often expressed in degrees. A wider FOV (e.g. ≥45°) covers more area, which is useful for close-proximity work or situational awareness. For example, a wide lens on a handheld thermal imager lets a technician scan a large section of a wall at once for insulation gaps. In contrast, a narrow FOV (telephoto lens, e.g. 6°–12°) is better for long-range observation – it magnifies distant targets (at the expense of covering a smaller scene). In a security context, a narrow FOV thermal module can detect humans at hundreds of meters but would be impractical for close quarters. Choose FOV based on target distance and coverage area: e.g. a 90° lens for a short-range situational camera, vs. a 12° lens for a long-range spotting scope. Some thermal cores offer multiple lens options – Teledyne FLIR Boson comes with horizontal FOV choices from ~4° up to ~92° depending on lens.

• Lens Focal Length: Often given in millimeters, it’s directly tied to FOV (for a given sensor size/pixel pitch). Common thermal module lenses include 4 mm (very wide), 7.5 mm, 13 mm, 19 mm, 35 mm, 50 mm, etc. A longer focal length yields a narrower FOV and higher optical magnification. Be mindful that if you change the default lens on a module, the device may need recalibration (for focus and flat-field correction) and the detection range will change. Some vendor data or range calculators can predict detection/recognition distances for various lens/sensor combos. For instance, a 19 mm lens on a 640×480 sensor (~12 µm pitch) might give ~32° horizontal FOV and detect a human ~500 m away, whereas a 50 mm lens (~12° FOV) could extend human detection beyond 1 km under the same conditions.

• F-Number (f/#): This is the lens focal length divided by its clear aperture. A lower f/# (e.g. f/1.0) lens is “faster,” meaning it admits more infrared radiation to the sensor, improving signal strength and potentially NETD. Most thermal modules use fast lenses (f/1.0 to f/1.3) to maximize sensitivity. A high f/# (say f/1.6 or f/1.8) would reduce the energy reaching the detector, effectively raising NETD (dimming the thermal image). When comparing modules, note if one uses an f/1.0 lens vs f/1.2 – the f/1.0 will have an edge in image brightness and sensitivity. However, lower f/# lenses are larger and more expensive, especially for longer focal lengths (e.g. a 100 mm f/1.0 germanium lens is very large and costly). There are practical limits: uncooled LWIR cameras beyond ~250 mm focal length become unwieldy and are often not offered because a cooled MWIR system would be more feasible for such ranges.

• Focus Mechanism: Simpler thermal modules have a fixed-focus lens (often set to see clearly from some minimum distance to infinity). This eliminates moving parts and makes integration easy, but you sacrifice the ability to adjust focus for very close targets. Higher-end modules or camera cores may support manual focus or even motorized autofocus. Manual or motor focus adds complexity (gears, motors, control interface) but allows the end-user to get a sharp image on targets at various distances – important for hand-held viewers or dual-use cameras. Determine if your application demands adjustable focus (e.g. a thermal inspection camera needing focus from 0.5 m to infinity) or if fixed-focus at a certain depth is acceptable (e.g. a rifle scope that might be parallax-optimized at long range). Many OEM modules come in both fixed-focus and adjustable versions.

Key takeaway: Optics should match your use-case. Ensure the module can be supplied with the desired lens, or that the vendor can provide lens-mount options if you plan to swap optics. Note that the lens and detector together determine the Instantaneous Field of View (IFOV) – essentially the ground sample distance per pixel at range – which in turn affects detection and identification criteria. If unsure, consult the module manufacturer’s lens charts or request their assistance in selecting the right lens for your detection range requirements.

Resolution and Frame Rate

Resolution (the pixel array size of the thermal sensor) and frame rate (the refresh rate of the thermal images) are two fundamental performance specs that often trade off with cost, data bandwidth, and in some cases, regulatory limits.

• Sensor Resolution: Common thermal module resolutions include 160×120 (QQVGA), 320×240 or 384×288 (QVGA), and 640×480 or 640×512 (VGA). Higher resolutions like 1024×768 (XGA) are emerging in high-end systems. A higher resolution sensor means each thermal image has more pixels, which provides more detail and the ability to resolve smaller or more distant objects. For example, a 640×480 thermal module can identify smaller features or individuals at range better than a 256×192 module, given the same lens. However, higher resolution sensors are more expensive and produce more data to process.

  It’s worth noting that in the current market for uncooled thermal OEM cores (as of mid-2020s), 384×288 and 640×512 are the dominant resolutions for performance-oriented applications. According to market data, about half of thermal sensors shipping for the hunting/outdoor market were QVGA (~384×288) and ~38% were VGA (640×480/512) in 2019 – indicating many OEM products opt for one of these. Lower resolution cores (<= 160×120) are used for compact or low-cost devices (like smartphone attachments or basic thermal viewers), while the latest 1024×768 or higher are still niche and costly. When choosing, consider the smallest target size and longest distance you need to resolve. If you need to see human-size objects at say 500 m, a 320×240 might only give a few pixels on target (detection but not recognition), whereas 640×480 could provide recognition-level detail at that distance. Some vendors provide Johnson’s criteria DRI (Detection, Recognition, Identification) distance estimates for their sensors with various lenses– use those as a guide.

• Pixel Pitch: Resolution goes hand-in-hand with pixel pitch (the size of each detector pixel, in micrometers). Modern uncooled sensors have pixel pitches of 12 µm or 17 µm typically, down from 25 µm a decade ago. Smaller pixel pitch allows more pixels in the same sensor area or a smaller sensor for the same resolution, which can reduce cost and lens size. For instance, FLIR’s Boson uses a 12 µm pitch detector in 640×512, enabling a very small module. One caveat: as pixel size shrinks, optics need to be sharper (higher Modulation Transfer Function) to fully utilize the resolution, and diffraction begins to limit performance when pixel size approaches the wavelength (for LWIR ~10 µm). Today’s 12 µm uncooled sensors are near that diffraction limit for LWIR, so going much smaller (e.g. 8 µm) will require optic innovations. Nonetheless, R&D trends show pixel sizes continuing to decrease – from ~45 µm in 1990 to 12 µm today, with 10 µm and even 5 µm pixel pitches demonstrated in research for the future. As an OEM, you mostly just need to know the pitch to understand lens compatibility (some datasheets specify that the lens is designed for a certain detector size and pitch). It also influences NETD and resolution trade-offs (more, smaller pixels can increase noise if not well engineered).

• Frame Rate: Thermal modules often come in two frame rate variants: a <9 Hz version (sometimes listed as 8.7 Hz or 7.5 Hz) that is exportable globally without stringent controls, and a 30 Hz or 60 Hz version which offers full-motion video but may require an export license for international sales. Higher frame rates are crucial for applications with motion – e.g. tracking fast-moving objects, UAV surveillance to avoid motion blur, or fluid thermal events. For a helmet-mounted or weapon-mounted scope, 30 Hz is typically the minimum for a smooth user experience (60 Hz even better). For a static sensor monitoring equipment, 9 Hz might be sufficient. Understand the export implications: Under US EAR (Export Administration Regulations), most uncooled thermal cameras above 9 Hz fall under ECCN 6A003 and need licensing for export, whereas <=9 Hz units are ECCN 6A993 and often No License Required to many countries. European export rules similarly distinguish high frame rates under the Wassenaar Arrangement (6A003.c in EU). If you plan to sell globally, you might opt for the 9 Hz model to ease logistics – but keep in mind that low frame rate video is choppy. Some OEMs mitigate this by doing image interpolation or shuttering tricks, but it’s not the same as true high frame rate capture.

In addition to frame frequency, check if the module supports windowing or analog video output frame rates. Some cores can output 60 Hz via analog (composite video) even if their digital interface is limited to 30 Hz. Others might allow window-of-interest readout at higher rates (e.g. a 640×480 sensor reading a 640×120 window at a faster rate). For most buyers, the standard options (9 Hz vs 30 Hz) are fixed. So choose based on application and compliance: e.g., a thermal sight for local law enforcement can use 30 Hz freely within your country, but a product aimed at international hunting markets might favor the 9 Hz variant to avoid export red tape.

Image Processing, Output Formats, and SDK Support

Another major decision is whether to get a “raw” thermal module vs one with integrated processing. Thermal sensors inherently output analog signals that need transformation into a human-viewable image (and sometimes temperature readings). Many OEM modules now include some level of on-board image signal processing (ISP) and offer a digital video output or data stream. Here are points to consider:

• Analog vs Digital Output: Legacy thermal cores provided analog video (NTSC/PAL) which can be convenient for simple integration (just feed to a display or video encoder). Modern modules more commonly output digital data (BT.656, parallel CMOS, USB, Ethernet, MIPI CSI-2, etc.). Ensure the module’s output interface is compatible with your system. For instance, if you want to connect directly to a Linux processing board, a USB or MIPI output is ideal, whereas if you are retrofitting into an analog CCTV infrastructure, you might need analog-out. Some cores do both (e.g. FLIR Tau 2 had analog and digital over different connectors). If the module outputs raw pixel data, you’ll need to apply false-color mapping and perhaps temperature calibration in your software.

• On-board Processing (ISP): Features like non-uniformity correction (NUC), auto gain control (AGC or histogram equalization for contrast), pseudo-color palettes, digital zoom, and even object detection may be handled on-board by smarter cores. Using a module with a built-in thermal image processor can significantly shorten development time – the module essentially behaves like a finished thermal camera, streaming a ready-to-use video feed. For example, an OEM camera module might output a crisp thermal image with dynamic range adjustments done automatically. In contrast, a basic sensor module could give you 14-bit raw thermal frames that look flat and require heavy processing to be useful. If you have thermal image processing expertise or need custom algorithms (perhaps to measure temperatures accurately, or custom image blending), a raw access module gives flexibility. Otherwise, leveraging the vendor’s ISP (if it meets your needs) can save you from reinventing the wheel of image optimization.

• Radiometric Data: Some thermal modules are radiometric, meaning they are calibrated to output actual temperature readings for each pixel (or at least allow conversion via known formulas). If your application involves temperature measurement (predictive maintenance, fire detection, body temperature screening, etc.), ensure the module is a radiometric type. Radiometric models often have a shutter for periodic recalibration and might output temperature in centigrade or as 16-bit data that your software converts. For instance, FLIR’s Lepton has a radiometric version that gives measurements. Many OEM cores (FLIR Boson, DRS Tamarisk, Seek Mosaic, etc.) offer radiometric modes. Non-radiometric ones are tuned just for imaging (higher contrast, not as accurate for temp). You can often identify radiometric models by looking for a stated accuracy spec (e.g. ±5 °C) and the presence of a calibration shutter.

• SDK and Software Integration: A critical consideration for OEMs is the software development kit (SDK) or API provided by the module manufacturer. A good SDK will let you control the camera’s features (like choosing color palettes, initiating a flat-field correction, adjusting gain settings, reading spot temperatures, etc.) and will come with libraries or code samples for common platforms (C/C++, Python, ROS, etc.). Check if the vendor provides SDK support for your target operating system (Windows, Linux, Android). For example, FLIR’s Boson SDK supports a USB Video Class mode and a separate command interface for camera control, and InfiRay or ULIS modules have their own SDKs. If you plan to do onboard image processing with your own algorithms (say for target recognition using AI), consider whether the module can output raw data in a convenient format and if it has enough interface bandwidth to stream that data to your processor. Some modules also support third-party middleware – e.g., drivers for GigE Vision, or integration with programs like LabVIEW, which might be relevant for industrial users.

• Output Format and Video Compression: Determine if you need the module to compress the video (for transmission) or if you will handle that. Some thermal cores on drones, for instance, output H.264/H.265 encoded video over IP to reduce bandwidth. This is not common on raw OEM modules (which usually output uncompressed frames), but a few integrated solutions might include an encoder. Most likely, if needed, you’ll add an encoder chip or use your processor’s capabilities. Just ensure you can get the frames out in a format suitable for that (e.g. 8-bit grayscale or YUV frames for encoding).

In summary, developers with limited thermal imaging experience should lean toward modules with strong ISP and SDK support, essentially treating it as a black-box thermal camera. Those with specific custom processing needs might choose a more bare-bones module where they can control every aspect (but then must implement NUC, image enhancement, etc., themselves).