Industrial Camera

What is an Industrial Camera

An industrial camera, more precisely a machine-vision camera, is a digital imaging device engineered for automated inspection, measurement, guidance, and identification on production lines and in instrumentation. It is the eye of an Industry 4.0 line: it turns light reflected from a part into structured pixel data that downstream vision software measures, classifies, decodes, or uses to steer a robot. While the optical front end resembles any digital camera, the engineering brief is entirely different, which is why an industrial camera is not interchangeable with a consumer or webcam device.

Three properties separate an industrial camera from a consumer camera. First, it exposes a deterministic, programmable interface: exposure time, gain, trigger source, and region of interest are all set by software and repeat identically frame after frame, so a measurement made today matches one made next year. Second, it delivers raw, uncompressed image data over a standardized machine-vision transport, avoiding the lossy JPEG compression that would corrupt a sub-pixel gauging result. Third, it is built for 24/7 duty cycles, hard external triggering, and precise synchronization with PLCs, encoders, and strobe illumination, so that an image is captured at the exact instant a part is in position under a pulse of light.

Functionally, the imaging chain is the same in every industrial camera: a lens focuses light onto a 2D area or 1D line photosensitive array; each pixel converts incoming photons to charge via the photoelectric effect; that charge is converted to a voltage, amplified, and digitized by an analog-to-digital converter (ADC). What differs between products is the geometry of the array, the sensor technology and shutter behavior, the spectral band the silicon or other material responds to, and the transport that carries the data out. Those four axes structure the rest of this guide.

Industrial cameras serve a broad set of tasks across the factory: surface and defect inspection, dimensional metrology and gauging, OCR/OCV and 1D/2D code reading, robot and vision guidance for pick-and-place, assembly verification, web inspection of paper, film, metal, and glass, electronics and semiconductor inspection, food and pharmaceutical sorting, and agriculture and recycling sorting using multispectral or SWIR imaging. Because the requirements of these tasks differ by orders of magnitude in speed, resolution, and spectral band, there is no single universal industrial camera; selection is the discipline of mapping an inspection task to a specific camera geometry, sensor, and interface.

The practical consequence for a procurement engineer is that the spec sheet, not the marketing photo, decides fitness for purpose. Resolution, sensor format, pixel size, frame or line rate, shutter type, bit depth, dynamic range, quantum efficiency, spectral range, interface, lens mount, trigger and I/O, and environmental rating each map to a concrete requirement of the inspection task. The chapters that follow define each of these and then assemble them into a repeatable selection sequence.

Industrial Camera Types

Industrial cameras are categorized first by scanning geometry, which is how the sensor builds an image, and second by spectral band or specialty, which is the part of the spectrum or the optical property the camera is built to capture. Getting the geometry right is the most consequential first decision, because it follows directly from whether the part is static or moving. The table below summarizes the main scanning geometries.

Area-scan cameras capture a full 2D frame in a single exposure, exactly like taking a digital photo. They are the general-purpose workhorse for discrete parts: presence and absence checks, dimensional gauging, OCR and barcode reading, and robot guidance. Because a single exposure freezes the whole field of view, area-scan is the right default whenever parts are indexed or static at the moment of capture, and it is by far the most common type in the field.

Line-scan cameras capture one (or a few) rows of pixels at a time and build the image as the object or the camera moves; the image height is created by relative motion synchronized to an encoder. This makes line-scan the natural fit for continuous web or roll goods such as paper, film, textiles, and metal or glass sheet, for cylindrical parts that are rotated past the sensor, and for very high-resolution inspection of large or fast-moving surfaces. TDI (Time Delay Integration) line-scan is a variant that sums charge over multiple stages to boost sensitivity, which is what enables low-light, high-speed inspection such as fast printing lines and web defect detection where a single-line exposure would be too dim.

3D cameras produce depth or height data rather than a flat intensity image. The technologies include laser triangulation (a laser profiler), structured light, stereo, and time-of-flight (ToF), each trading resolution, speed, range, and cost differently. Smart cameras integrate the sensor, processor, and I/O in one housing and run the vision software onboard, removing the need for an external PC; they trade flexibility for simpler deployment, and suit standalone checks. Embedded or board-level cameras are bare PCB modules intended for OEM integration into instruments and devices, where the device builder supplies the housing, optics, and processing.

Beyond geometry, cameras are also classified by spectral band or specialty. Visible mono or color is the standard. NIR (near-infrared, ~750-1000 nm) uses silicon sensors with NIR-enhanced quantum efficiency. SWIR (short-wave infrared, ~900-1700 nm) uses InGaAs sensors, for example the Sony SenSWIR IMX99x family, for moisture, silicon-wafer, hot-glass, and sorting applications. UV (~200-400 nm) reveals surface defects and fluorescence. Thermal/LWIR uses microbolometer arrays for temperature and heat imaging. Polarization cameras place on-chip 4-direction polarizer arrays over the sensor, for example the Sony IMX250MZR Polarsens, to reveal stress, scratches, and glare. Finally, event-based and high-dynamic-range (HDR) cameras occupy a niche for very high-speed or wide-dynamic-range scenes.

Operating Principle and Sensor Technology

Inside every industrial camera, light is focused by a lens onto a 2D (area) or 1D (line) photosensitive array. Each pixel converts photons to charge through the photoelectric effect; the charge is converted to a voltage, amplified, and digitized by an ADC. The differences that drive selection and image quality come down to four technology axes: CMOS versus CCD, global versus rolling shutter, mono versus color, and back-side-illuminated or stacked sensor construction. The table below compares the two shutter behaviors and the two color paths that most affect a buying decision.

CMOS versus CCD. CMOS is now the industry standard, having displaced CCD for almost all machine-vision use. A CMOS sensor reads each pixel through on-chip circuitry, which gives higher frame rates, lower power, and lower cost. CCD, which transfers charge to a common readout, historically offered superior uniformity and low noise, but it is largely obsolete and reaching end-of-life for new designs. Modern global-shutter CMOS families such as Sony Pregius and onsemi sensors have closed the legacy image-quality gap, so there is rarely an engineering reason to specify CCD for a new system.

Global shutter versus rolling shutter. A global shutter exposes all pixels simultaneously and then reads them out, so there is no motion distortion; this is essential for moving parts, fast lines, and pulsed or strobe lighting. A rolling shutter exposes and reads row by row, which is cheaper and often allows higher resolution, but it produces skew, wobble, and banding on moving objects. Sony Pregius, and the BSI-stacked Pregius S, is the dominant industrial global-shutter CMOS family, which is why so many area-scan cameras list a Pregius generation in their spec.

Mono versus color. Mono sensors have higher effective resolution and light sensitivity because they carry no Bayer color filter array; color sensors use a Bayer CFA and demosaicing to reconstruct color, which costs resolution and light. The rule is to choose mono for gauging and defect work, and color only when color is the discriminating feature of the inspection.

BSI and stacked construction. Back-side-illuminated (BSI) and stacked sensors, such as Pregius S, route wiring away from the light path so the pixel can be shrunk while still raising quantum efficiency. This is how a modern sensor packs more, smaller pixels into a given format without surrendering the sensitivity that small pixels would otherwise lose.

Photo: Phiarc, CC BY-SA 4.0, via Wikimedia Commons

Sensor Materials and Spectral Matching

The material of the sensor decides which part of the spectrum the camera can see, so spectral matching is fundamentally a materials decision. Choosing the wrong material is not a tuning problem that better illumination can fix; if the sensor material does not respond to the wavelength your feature emits or reflects, the feature is simply invisible. The table below maps sensor materials to the spectral bands they cover.

Silicon CMOS and CCD respond across the visible and near-infrared, roughly 350-1000 nm, but their sensitivity falls off above about 1000 nm. This is why a standard silicon camera, even an NIR-enhanced one, cannot reach into the short-wave infrared; the physics of silicon sets the ceiling. For the great majority of inspection tasks that live in the visible and NIR, silicon is the obvious and economical choice.

InGaAs sensors cover the SWIR band, roughly 900-1700 nm, and are used in SWIR cameras for through-material imaging and thermal-glass imaging, such as seeing moisture, inspecting silicon wafers, imaging hot glass, and sorting materials by their SWIR signature. InGaAs sensors are often thermoelectrically (TEC) cooled to cut dark current, which improves the signal-to-noise ratio in this demanding band.

Microbolometer arrays, built from vanadium oxide (VOx) or amorphous silicon, provide uncooled long-wave infrared (LWIR) thermal imaging in the 8-14 µm band. These cameras measure emitted heat rather than reflected light, which is what makes them suited to temperature and heat imaging. On-chip polarizer and micro-lens arrays add an optical filtering layer directly over the silicon to enable polarization and color filtering, the technology behind polarization cameras.

The camera housing is itself an engineering choice. Industrial camera housings are typically anodized aluminum for heat dissipation and rigidity, which keeps the sensor temperature stable and the optics in alignment. Ruggedized models add gasketed, sealed enclosures and screw-lock connectors to achieve IP-rated protection, so the same sensor can be deployed in a clean lab or in a washdown food line by changing the enclosure rather than the imager.

Key Specification Parameters

Reading an industrial-camera spec sheet is a core skill for purchasing engineers, because the same camera can look very different across vendor datasheets. The parameters that truly drive selection are resolution, sensor optical format, pixel size, frame or line rate, shutter type, bit depth, dynamic range, quantum efficiency, the EMVA 1288 sensitivity metrics, spectral range, interface, lens mount and flange distance, trigger and I/O, and environmental rating. The table below summarizes the parameters, with units and typical ranges, before each is explained.

Resolution is the total pixel count, from 0.3 MP (VGA) up to 24.5+ MP for area-scan, while line-scan is rated as 2k, 4k, 8k, or 16k pixels per line. The required resolution comes from the smallest defect or feature divided by the pixels-per-feature you need to allocate across the field of view. Sensor optical format is the diagonal class: 1/4", 1/3", 1/2.3", 1/1.8", 2/3", 1", 1.1", 4/3 (MFT), APS-C, and full-frame. The practical maximum for C-mount is about 1.1" (around 17.6 mm diagonal), and the format must match the lens image circle or the corners will vignette.

Pixel size is typically about 1.45-5.86 µm. The Sony Pregius generations illustrate the trend: 5.86 µm first-generation, 3.45 µm second-generation ("3.4 series"), 4.5 µm third-generation, and 2.74 µm Pregius S ("2.7 series"). Larger pixels mean more sensitivity and full-well capacity, while smaller pixels mean higher resolution per unit area, so pixel size is a direct sensitivity-versus-resolution trade. Frame rate (area) and line rate (line-scan) set throughput: area-scan reaches tens to thousands of fps at reduced ROI, and line-scan reaches tens to hundreds of kHz lines per second. Shutter type, global versus rolling, was covered in Chapter 3 and governs motion distortion and strobe compatibility.

Bit depth (ADC) is commonly 8, 10, or 12 bit, with some 14- or 16-bit cameras for scientific or HDR work; higher bit depth gives finer grayscale gradation. Dynamic range is the ratio of saturation capacity to read noise, stated in dB, typically around 60-75 dB for CMOS, with HDR modes reaching higher; it is defined per EMVA 1288. Quantum efficiency (QE) is the percentage of incident photons converted to electrons at a given wavelength, with peak QE often around 60-90% for modern BSI mono sensors. Alongside these, the core EMVA 1288 sensitivity metrics are saturation capacity / full well (e-), temporal dark noise / read noise (e-), SNR (dB), and absolute sensitivity threshold (photons), while DSNU and PRNU quantify dark-signal and photo-response nonuniformity (spatial noise). Spectral range is the band the sensor responds to, as covered in Chapter 4.

Interface sets bandwidth, cable length, power, and multi-camera capability, and is detailed in the standards box below. Lens mount and flange distance determine which optics fit: C-mount uses a 1"-32 UN thread with a 17.526 mm flange; CS-mount uses the same thread with a 12.526 mm flange, exactly 5 mm shorter, so a C-mount lens fits a CS body with a 5 mm spacer but not the reverse. M12/S-mount serves small formats up to about 1/2.3", while larger metric mounts M42/M58 and the F-mount serve large sensors and telecentric and metrology optics. Trigger, I/O, and synchronization cover hardware trigger latency and jitter, opto-isolated GPIO, encoder input for line-scan, and PTP/IEEE 1588 time sync. Environmental specs cover operating temperature range, vibration and shock rating, and ingress protection (for example IP67 means dust-tight plus immersion to 1 m for 30 minutes), while power and power-over-cable options are PoE over GigE or bus power over USB3 versus an external supply.

Governing standards. Interface and transport: GigE Vision runs machine vision over Gigabit Ethernet (first released May 2006, revised v1.2 in 2010, v2.0 in 2011, v2.1 in 2018), in 1 GigE (~1 Gbps / ~100+ MB/s), 5 GigE, and 10 GigE variants, with cable up to 100 m over standard Ethernet, PoE support, and easy multi-camera; it is certified by the Association for Advancing Automation (A3). USB3 Vision runs over USB 3.x (USB 3.0 = 5 Gbps, ~380 MB/s usable) with short cable reach (passive ≈ 3-5 m recommended, up to 8 m per Basler), bus-powered and simple for a single camera, also certified by A3. Camera Link is a serial point-to-point link with a frame grabber, in configurations Base ≈ 2.04 Gbit/s (255 MB/s), Medium ≈ 4.08 Gbit/s (510 MB/s), Full ≈ 5.44 Gbit/s (680 MB/s), and 80-bit Deca ≈ 6.8 Gbit/s (max ~850 MB/s), with cable up to 10 m for Base at any clock rate, dropping to roughly 5 m at the 85 MHz clock for Medium, Full, and Deca. Camera Link HS (CLHS) is a scalable high-speed standard with CX4 copper up to ~15 m (~5 Gbps/lane) and fiber for very long reach, supporting forward error correction. CoaXPress (CXP) runs high speed over coax; CXP 2.0 (administered via JIIA) adds CXP-10 (10 Gbps) and CXP-12 (12.5 Gbit/s per coax), uses 1/2/4/8 coax cables (a 4-link CXP-12 cable carries ~4 GB/s), reaches up to ~40 m on a single coax, and provides power-over-coax (up to 13 W per cable) plus a low-latency uplink (20.83 Mbps for CXP-1 to CXP-6, doubled to 41.66 Mbps at CXP-10/CXP-12, with an optional 6.25 Gbit/s high-speed uplink in CXP 2.0). Software and characterization: GenICam (Generic Interface for Cameras), administered by the EMVA, decouples the application API from the physical interface, with components GenApi (feature configuration), GenTL (generic transport layer for enumeration, buffering, and image transfer), SFNC (Standard Features Naming Convention), and PFNC (pixel-format naming). EMVA 1288 is the standard for objective characterization of sensors and cameras, defining unified measurement of responsivity, QE, saturation capacity, SNR, dynamic range, temporal noise, linearity, dark current, and spatial nonuniformity (DSNU/PRNU); a datasheet is EMVA 1288-compliant only if all mandatory measurements from at least one camera are reported (versions 3.x and the newer 4.0 are in field use). For standards-body context, GigE Vision, USB3 Vision, Camera Link, and Camera Link HS are stewarded by the AIA/A3, CoaXPress by the JIIA, and GenICam and EMVA 1288 by the EMVA, with the three associations coordinating via the international G3 initiative.

Selection Decision Factors

To turn the preceding five chapters into a specific model, follow the decision sequence below. Most selection mistakes come not from a single wrong step but from deciding the wrong thing first; resolving the task definition before the interface, and the interface before the housing, keeps the chain consistent. These ten steps can serve as a fixed RFQ template.

1. Define the inspection task: Fix the smallest feature, the field of view, and the accuracy, which together give the required resolution. Allocate pixels-per-feature, typically at least 3 px per smallest feature or defect.
2. Object motion: Static or indexed parts call for area-scan; continuous web, cylindrical, or very large surfaces call for line-scan with an encoder. Fast motion or strobe makes global shutter mandatory.
3. Speed: Throughput sets the required frame or line rate, which, together with resolution times bit depth, sets the data bandwidth and therefore dictates the interface choice.
4. Interface: Balance distance, bandwidth, multi-camera, and cost. USB3 is short, cheap, single-camera; GigE/PoE gives long reach (100 m) and easy multi-camera; CoaXPress or Camera Link HS gives the highest bandwidth with a frame grabber and long coax or fiber.
5. Color versus mono, or spectral band: Mono for gauging and defect sensitivity; color only if color discriminates; NIR/SWIR/UV/thermal/polarization for non-visible features.
6. Sensitivity and light budget: Weigh pixel size, QE, full well, read noise (EMVA 1288), and dynamic range, then pair them with the illumination and exposure budget.
7. Optics: Match the sensor format to the lens image circle and the mount (C/CS/M12/F/M42/M58); the lens resolution must match the pixel pitch, and metrology may require telecentric optics.
8. Environment: Confirm temperature, vibration, and IP rating (IP67 for washdown or dust) plus connector locking.
9. Software and ecosystem: Require GenICam/GenTL compliance for vendor-neutral integration, and verify SDK and third-party library support such as Halcon, OpenCV, and vendor SDKs.
10. Cost and lifecycle: Total the unit cost, any frame grabber requirement, cabling, and long-term availability.

One last dimension that buyers often overlook is the breadth and maturity of the manufacturer. Established suppliers span the geometry, interface, and spectral choices above so that a line can standardize on one ecosystem. Basler AG (Germany) offers the ace, ace 2, boost, dart, and blaze families across GigE, USB3, and CXP, plus a visSWIR line. Teledyne covers DALSA line-scan and TDI (Genie, Linea, Falcon), Teledyne FLIR machine vision and thermal, Lumenera, and Photometrics/Princeton scientific imaging. Cognex Corporation (USA) makes In-Sight smart cameras and DataMan ID readers and acquired Moritex optics in 2023. Allied Vision (Germany, TKH Group) offers Alvium, Mako, Manta, and Goldeye (SWIR). Sony Semiconductor Solutions is the sensor maker whose Pregius, Pregius S, SenSWIR, Polarsens, and Starvis families power most industrial cameras, alongside onsemi as a major CMOS supplier (XGS, Python). HIKROBOT / Hikvision Machine Vision (China) ships high-volume area, line-scan, and smart cameras; Baumer Group (Switzerland) makes VeriSens smart cameras and UV/NIR/SWIR/polarization cameras; IDS Imaging Development Systems (Germany) makes uEye and IDS NXT AI cameras; JAI (Denmark) makes prism, multi-spectral, line-scan, and high-end area-scan; and LUCID Vision Labs (Canada) makes the IP67 Triton and Atlas GigE/PoE, SWIR, and polarization cameras. Specialists include Chromasens (line-scan), Omron (vision systems), Photonfocus, e-con Systems / Vadzo (embedded), and Euresys / Matrox / KAYA (frame grabbers and CXP).