Image Sensor Selection for Machine Vision: Resolution, Shutter Type, and Lens Pairing by Application
This guide matches resolution, shutter, spectral response, and interface bandwidth to the inspection task, not the megapixel count.
Select an image sensor by working backward from the inspection task. Required feature resolution sets pixel pitch, and motion profile sets shutter type. Illumination strategy sets spectral response, and process speed sets frame rate against interface bandwidth. The lens comes after: it must cover the sensor format and resolve the pixel pitch at the working F-number, or the sensor's resolution is wasted.
How Much Resolution Does the Inspection Task Need?
Resolution requirement comes from the smallest feature you must detect, not an arbitrary megapixel target. A common starting point is 3 to 5 pixels across the feature, though the number you need depends on feature contrast, lens MTF, sensor noise, illumination, and the detection algorithm. A 50µm defect sampled at 3 pixels needs roughly 17µm per pixel in object space, which sets the field of view a given sensor resolution can cover at your working distance.
Work the chain in order. Feature size sets the sampling requirement, and sampling plus field of view sets the pixel count. Pixel count plus format then sets pixel pitch. Jumping to "more megapixels" oversizes the optics, the interface, and the processing budget without improving detection.
A 25MP sensor sampling a feature that only needed 5MP wastes interface bandwidth and per-unit cost with no accuracy benefit. Undersizing is the more expensive mistake to discover late: a sensor that cannot resolve the required feature forces a full camera and lens respin.
Calculate required pixel pitch before comparing sensor part numbers. Two sensors with the same megapixel count but different formats have different pixel pitches, and pixel pitch, with the magnification that projects it onto the part and the lens MTF it demands, decides whether the system resolves the feature; megapixel count alone settles none of that. Commonlands sizes each lens it recommends to the sensor's pixel pitch, not only its format. See spatial resolution in machine vision for the full sampling-to-lens-MTF chain.
Global Shutter vs Rolling Shutter for Machine Vision
Global shutter starts and ends integration for every pixel together, giving all rows one common exposure interval; the capture is not instantaneous, and motion blur still scales with exposure time. Rolling shutter exposes rows sequentially over a readout period lasting microseconds to milliseconds. For a stationary scene, the difference is invisible. For anything that moves relative to the camera (the object, the camera, or both), rolling shutter introduces geometric distortion because the scene changed between when the first row and the last row were sampled.
Strobed illumination needs care on rolling shutter. If the strobe fires while only part of the array is integrating, only those rows record the flash, leaving a bright band. Under strobe-dominated illumination with negligible ambient, the pulse width sets effective exposure and motion blur; otherwise the programmed exposure accumulates ambient smear too. The exposure setting only has to open a window in which every row integrates at once.
Whether that window exists depends on the sensor timing: an exposure longer than the full readout creates one, and parts with a global reset mode start all rows together instead. Fire the short pulse inside that window to freeze motion.
The tradeoff comes down to motion and the geometric error the task tolerates. Global shutter is the safer default when the camera or object moves during exposure (conveyors, robotics, pick-and-place, strobed illumination, precision measurement): a 200mm/s conveyor moves 0.2mm during a 1ms readout, a real error for a 10µm feature.
Rolling shutter fits when nothing moves during readout (document capture, static label reading, kiosk scanning, cost-sensitive embedded modules). It can still work with motion when readout is short, skew stays inside the task tolerance, or a strobe freezes the scene; speed, motion direction, triggering, correction, cost, and noise move that line.
| Factor | Global shutter | Rolling shutter |
|---|---|---|
| Motion tolerance | No readout skew or wobble; motion blur still depends on exposure time | Skew and wobble scale with speed and readout time |
| Strobed illumination | Compatible with short pulses; pulse width sets effective exposure when strobe light dominates ambient | Short pulses work only inside a window where all rows integrate together (exposure longer than readout, or a global reset mode, per the sensor timing); otherwise banding |
| Typical cost | Higher, since a storage capacitor per pixel adds die area | Lower at a given resolution and pixel pitch |
| Low-light sensitivity | Often slightly lower QE at equal pixel pitch (older designs) | Often higher QE and lower noise at equal pixel pitch |
| Common mount pairing | C-mount, larger formats, adjustable iris | M12, compact embedded modules |
| Typical sensors | IMX253, IMX264, IMX568, AR0234 | IMX477, OV5640, IMX415, IMX678 |
Shutter type also correlates with lens mount: global shutter cameras in higher-accuracy industrial systems commonly pair with C-mount lenses, whose adjustable iris ring gives depth-of-field control (practical because illumination is usually programmatically controlled). Rolling shutter sensors dominate compact embedded modules paired with M12 lenses for size and weight. The Commonlands M12 vs C-mount vs CS-mount guide covers the full tradeoff.
Pixel Size vs Lens Resolving Power
Pixel pitch is the center-to-center distance between adjacent pixels. Smaller pixels pack more resolution into a given sensor format, but each pixel captures less light and the lens must deliver higher contrast at finer spatial frequencies to resolve detail at pixel scale. A sensor's resolution is only as good as the lens resolving it: pair a small-pixel sensor with a lens specified for a lower-resolution sensor and you get soft detail no amount of sensor resolution recovers.
Diffraction caps lens sharpness regardless of lens quality, and it is gradual: contrast falls progressively with spatial frequency, reaching zero only at the optical cutoff, and the falloff grows with F# and wavelength.
Commonlands specifies each lens for a target sensor resolution and pixel pitch; the rating names the sensor class rather than guaranteeing resolved pixels, so confirm the pairing against the lens's measured MTF data at your pixel pitch, working aperture, and field position. The aperture-versus-depth-of-field tradeoff this creates is covered in full in f-number in machine vision.
NIR Sensitivity and Illumination Strategy
Sensor spectral response should follow the illumination strategy, not the other way around. Standard silicon CMOS sensors retain meaningful quantum efficiency into the near-infrared, typically out to 1000–1100nm, but most machine vision cameras ship with an IR-cut filter installed to preserve visible-light color accuracy.
If your system illuminates with 850nm or 940nm LEDs (common for covert lighting, low-visible-light environments, or combined day and night operation), remove the IR-cut filter or specify a NIR-optimized variant. Then confirm the sensor's QE curve at your chosen wavelength rather than assuming uniform NIR sensitivity across parts.
At 850nm, most silicon sensors keep higher QE than at 940nm, at the cost of a faint visible red glow that is sometimes undesirable in public-facing installations. 940nm light is invisible to the eye, but silicon QE there is often around half, with the exact ratio sensor-specific, so it needs brighter illumination or a faster aperture to compensate. Match sensor, filter, and illuminator wavelength as one decision, not three: a NIR-sensitive sensor behind a standard visible bandpass filter gains nothing from the illuminator.
Confirm image circle and lens coating compatibility with your NIR band. See bandpass filter machine vision for lens-side filter selection, and browse the Commonlands filter collection for stocked bandpass and IR-cut options.
Frame Rate vs Interface Bandwidth
Frame rate is a function of sensor resolution and interface bandwidth together, not sensor speed alone. A 25MP sensor over GigE Vision maxes out below 10 fps; the same sensor over CoaXPress at 25 Gbps sustains 45 fps or more. A high-resolution sensor chosen without confirming the interface meets the resolution spec but misses the throughput spec.
| Interface | Typical bandwidth | Cable length | Best fit |
|---|---|---|---|
| USB3 Vision | 380 MB/s | ~5m | Benchtop and lab systems, simple integration |
| GigE Vision | 125 MB/s | 100m (standard Ethernet) | Lower-resolution or lower-frame-rate systems, long cable runs, low per-port cost |
| 10GigE Vision | 1.25 GB/s | Long runs, more expensive switching | High-resolution systems needing longer cable runs than USB3 |
| CoaXPress | up to 12.5 Gbps (~1.56 GB/s) per lane | Multi-lane configs available | Highest resolution and highest frame rate; requires a specialized frame grabber |
Commonlands engineering can size the lens once the sensor and interface are fixed.
Sensor Format and Lens Coverage
Sensor format is the physical size of the imaging area. The diagonal measurement sets how large a lens image circle you need. If the image circle is smaller than the sensor diagonal, the corners fall outside the rated coverage and receive little to no light, producing vignetting whose severity depends on the lens and how the image circle is defined.
Larger sensor formats capture a wider field of view at a given focal length, or let you use a longer focal length to hold the same field of view with a shallower depth of field. See sensor size and lens compatibility for the format-to-dimension reference and the vignetting math.
| Sensor format | Example sensors | Pixel pitch | Typical MP | Commonlands lens |
|---|---|---|---|---|
| 1/2.8" | IMX327 | 2.9µm | 2–5 MP | M12 lenses |
| 1/4" | OV5640 | 1.4µm | 5 MP | M12 lenses |
| 1/2.3" | IMX477 | 1.55µm | 12 MP | M12 lenses |
| 1/1.2" | IMX585 | 2.9µm | 8 MP | C-mount lenses |
| 2/3" | IMX264 | 3.45µm | 5 MP | C-mount lenses |
| 1.1" | IMX253 | 3.45µm | 12 MP | C-mount lenses |
| 1/2.6" | AR0234 | 3.0µm | 2.3 MP | M12 lenses |
| 1/1.8" | IMX547 | 2.74µm | 5 MP | C-mount lenses |
| 1.1"–1.2" | IMX532, GMAX0505 | 2.5–2.74µm | 16–25 MP | C-mount lenses |
A lens rated for a 2/3" sensor leaves the corners of a 1.1" sensor well outside its rated image circle, so severe corner shading is the expected outcome at any aperture; the exact falloff depends on the image-circle definition, aperture, and conjugate, so judge it from relative illumination data rather than the format label alone. Stopping down cannot fix a coverage mismatch: the shortfall is geometric, not a depth-of-field effect.
Always match or exceed the sensor format with the lens specification. Oversizing the lens format covers the diagonal, but coverage alone does not qualify the pairing: check CRA against the sensor's microlens and filter stack, MTF and relative illumination at your pixel pitch, mechanical clearance, and the design conjugate, and expect some added cost.
Recommended Lenses by Sensor Format
For machine vision sensors up to the 1.1 inch format, Commonlands stocks a matched lens in each band below. Each row lists the stock lens whose rated format and resolution meet or exceed the sensor, taken from the sensor format table above. Coverage comes from each lens's rated image format, not from field-of-view math done on this page.
| Sensor format band | Top pick | Mount & EFL | Why it fits |
|---|---|---|---|
| Embedded, up to 1/1.7" (OX08B40, AR0821 class) | CIL059 6mm low-distortion M12 | M12, 5.9mm | Rated up to 1/1.7", 4–6MP at F/1.7. The fast aperture suits low-light embedded modules. |
| Embedded and sealed, up to 1/1.8" | CIL034 IP67 3.2mm M12 | M12, 3.25mm | Rated up to 1/1.8", 5–10MP variants. IP67 sealing is specific to this SKU, not a property of all M12 lenses. |
| 1.1" 12MP industrial (IMX253, IMX304) | CIL508 8mm C-mount | C-mount, 8mm | Rated for 1.1" 12MP at F/2.4 with an adjustable iris. Wider field than the CIL512 at the same working distance. |
| 1.1" 12MP industrial, longer reach | CIL512 12mm C-mount | C-mount, 12mm | Same 1.1" 12MP coverage as the CIL508, with a longer working distance for a tighter field. |
| 1.1"–1.2" high-resolution, 20–25MP (GMAX0505, IMX541) | CIL542 12mm 25MP C-mount | C-mount, 12mm | Rated for 2.5µm pixel pitch at 25MP-class sensors; verify with its measured MTF data. Matched to high-MTF small-pixel sensors where an underrated lens wastes detail. |
Sensors larger than the 1.1 to 1.2 inch formats here, such as 35mm-format line-scan sensors, need F-mount or M42 class optics from vendors like Schneider or Zeiss. Commonlands does not stock that class.
Confirm coverage on your sensor at your working distance with the field of view calculator, and check the sensor diagonal against each lens image circle before you commit.
Frequently Asked Questions
How do I choose the right image sensor for machine vision?
Start with the inspection task, not the datasheet. Define the smallest feature you must resolve, whether the object or camera moves during exposure, the illumination band (visible or NIR), and the frame rate the process requires. Those four answers narrow the sensor list before you compare a single megapixel figure.
What is the difference between global shutter and rolling shutter?
Global shutter gives every pixel one common integration interval: all rows start and stop exposing together, though the capture still has finite duration, so motion blur depends on exposure time. Rolling shutter exposes rows sequentially over a readout period lasting microseconds to milliseconds. For a static scene the difference is invisible. For anything moving relative to the camera, rolling shutter introduces skew, wobble, or flash banding.
What pixel pitch do I need for machine vision?
A common starting point is 3 to 5 pixels across your smallest feature, adjusted for contrast, lens MTF, sensor noise, and your detection algorithm; from there, work backward through your magnification to the required pixel pitch at the sensor. Solve aperture and pixel pitch together, not independently.
Do I need a NIR-sensitive sensor for my application?
Choose a NIR-sensitive sensor and remove or bypass the IR-cut filter when your illumination uses 850nm or 940nm LEDs, common for low-visible-light environments or combined day and night operation. Confirm the sensor's QE curve at your wavelength rather than assuming uniform NIR sensitivity.
How do I match a Commonlands lens to my sensor?
Identify your sensor format and pixel pitch, then choose a Commonlands lens rated for that format or larger.
Use the Commonlands field of view calculator to confirm coverage at your working distance, and the depth of field calculator to check depth of field at your aperture. Contact Commonlands engineering if you are still unsure.
Need Help Matching a Lens to Your Sensor?
Commonlands engineering can recommend the right lens for your sensor format, pixel pitch, shutter type, and working distance before you commit to hardware.