Computer Vision lenses: FOV, WD and matching the sensor

What a Computer Vision lens is

A Computer Vision lens is an industrial optic that forms a sharp image of your part onto a camera sensor at a defined working distance and field of view. Unlike a phone camera lens, it is designed for fixed geometry, repeatable focus, low distortion (when you measure), and continuous duty next to robots, conveyors and strobes.

The lens is half of the optical path. The other half is the sensor inside your camera. Megapixels without a matching lens image circle, resolution class and focal length are empty promises. Teams that pick the camera first and “find a lens later” often rediscover that later means redesigning the station height, light and FOV.

Sedeco groups industrial optics under the lenses hub. This knowledge page teaches the language so product pages make sense: FOV, WD, magnification, mount, sensor format and distortion. We do not invent SKU focal lengths or MTF curves here, those live on current product datasheets.

Field of view and working distance

Two numbers dominate almost every lens conversation: field of view (FOV) and working distance (WD).

Field of view

FOV is the width and height of the scene that fills the sensor, for example 120 mm × 90 mm on a conveyor window. It must cover the part plus placement tolerance (and sometimes multiple parts). Too small and edges clip when the robot misses by a few millimetres; too large and you waste pixels, reducing spatial resolution on the feature that matters.

Working distance

WD is the distance from the lens (often the front element or a vendor-defined flange reference, check the datasheet definition) to the object plane. Machine builders care because WD must clear tooling, light housings and safety guards. Short WD packs magnification; long WD needs longer focal lengths or different optics for the same FOV.

How they relate

For a thin-lens approximation: FOV ≈ sensor size × (WD / focal length) in the appropriate units, more precisely, magnification and conjugate distances govern the mapping. In practice you either use a vendor FOV calculator, a lens selection chart, or the simple rule: larger sensor or shorter focal length → larger FOV at a given WD; longer focal length → smaller FOV (telephoto behaviour) at the same WD.

Write FOV and WD on a napkin before you open a catalogue. Then choose focal length and sensor format together. Pair this with an area scan camera when the station is a classic 2D inspection cell.

Magnification and spatial sampling

Magnification (m) is the ratio of image size on the sensor to object size in the world. At m = 0.1, a 10 mm feature spans 1 mm on the sensor. At m = 1 (1:1), object and image are the same size, common in precision gauging with telecentric lenses.

In practice you usually care less about “m” as a number and more about µm per pixel: how many micrometres of the part each sensor pixel represents. That value is FOV width divided by the number of pixels used across that width. Your smallest critical feature should span several pixels after optical MTF and lighting are included, not only in a calculation.

Example: FOV width 80 mm on a 2448-pixel-wide sensor → about 32.7 µm/pixel. A 0.2 mm defect spans about six pixels, often workable if contrast is good. Stretch the FOV to 160 mm without changing the sensor and you have half the sampling; the same defect may become unreliable.

C-mount, sensor format and image circle

C-mount is the workhorse thread and flange standard for many industrial area scan cameras: 1 inch–32 tpi thread and a defined flange focal distance. CS-mount is similar but with a shorter flange distance, adapters exist, but you must know which your camera expects. Larger sensors and specialized cameras may use F-mount, M42, M58 or proprietary mounts.

Sensor format (for example 1/2.9", 1/1.8", 2/3", 1", or larger) describes the diagonal of the active sensor area. Every lens has an image circle, the diameter of the sharp illuminated circle it projects. The image circle must be at least as large as the sensor diagonal. Undersizing causes vignetting (dark corners) and soft edges; “a C-mount lens” is not automatically safe for every C-mount camera.

When you upgrade from a small sensor to a larger one “for more megapixels,” re-check the lens. Many kit lenses that looked fine on 1/1.8" vignette on 1". Browse formats and families on the lenses product hub alongside the camera datasheet.

Distortion, telecentricity and measurement

Distortion bends straight lines in the image, barrel or pincushion shapes are common. For presence checks and code reading, a few percent may be acceptable. For dimensional gauging in pixels-to-millimetres, distortion becomes error unless you calibrate it out or choose low-distortion / telecentric optics.

Entocentric (ordinary) lenses change magnification with object distance and show perspective: objects farther from the optical axis look smaller. That is fine for many inspections. Telecentric lenses keep magnification nearly constant over a depth range and reduce perspective error, preferred for precise 2D metrology of edges and diameters when budget allows.

If your acceptance criterion is “±0.05 mm on a stamped blank,” discuss distortion and telecentric options early. If the job is “is the label present?”, a standard fixed focal length C-mount lens with good lighting usually wins on cost and complexity.

Optical resolution, aperture and depth of field

Sensor megapixels are only useful if the lens can deliver the spatial frequencies those pixels sample. Lens datasheets speak in line pairs per millimetre (lp/mm) or MTF charts. A cheap lens on a high-resolution sensor softens fine edges, you paid for pixels you cannot resolve.

Aperture (f-number) trades light and depth of field against diffraction. Wide open (small f-number) gathers light and may reduce depth of field and increase aberration at the edges. Stopping down deepens focus but can soften via diffraction and requires more light or longer exposure, which invites motion blur unless you strobe. Computer Vision lighting and lens aperture are one budget; see the lighting hub.

Depth of field must cover part height variation and presentation tolerance. Tall objects or floating parts on a belt may need smaller apertures, telecentric designs, or multiple views, not only “more megapixels.”

Common industrial lens types

Liquid lenses and motorized focus appear on smart cameras and code readers when WD varies. For fixed PC-based stations, a locked manual focus with threadlocker after commissioning remains common and robust.

Worked example: choosing FOV and focal length

Suppose you inspect a plastic cover that is 70 mm wide. Placement can wander ±10 mm, so you design a 90 mm FOV width. The critical flash is 0.3 mm; you want ≥5 pixels across it → ≤0.06 mm/pixel. Across 90 mm you need about 1500 pixels of width, a ~2 MP class sensor can work if optics are sharp and the FOV is tightly controlled.

Station height allows a WD of about 300 mm from lens front to part. With a sensor active width of about 7 mm (illustrative 1/1.8"-class), magnification ≈ 7 / 90 ≈ 0.078. Focal length falls near f ≈ m × WD / (1 + m) in a simple conjugate model, on the order of a mid-teens to ~20 mm class lens depending on exact conjugates and mount. You then verify with the vendor chart for your exact sensor and confirm image circle coverage.

Numbers on a whiteboard beat shopping by focal-length habit. Validate with real samples under real lighting; Sedeco’s feasibility approach exists for exactly this step.

Specs that matter on a lens datasheet

• Focal length, primary control of FOV at a given WD and sensor size.
• Supported sensor format / image circle, must cover your chip diagonal.
• Mount, C/CS or other; adapters change flange distance.
• Resolution class, matched to megapixel and pixel size of the camera.
• Aperture range, light vs DOF; lockable iris preferred in factories.
• Distortion, stated %; critical for gauging.
• Minimum object distance / WD range, whether focus reaches your plane.
• Filter thread, for polarizers, bandpass filters, protection windows.

Filters, windows and mechanical accessories

The optical path rarely ends at the last glass element. Polarizers cut specular glare on plastics and metals, often the difference between a readable Data Matrix and a washed-out blob. Bandpass or colour filters can isolate LED wavelengths or suppress ambient sodium/LED plant lighting. Protection windows and IP-rated housings keep coolant mist off the front element; they also add reflections and can shift focus slightly, so commission with the window in place.

Extension tubes and close-up adapters increase magnification for small parts but reduce WD and can degrade edge performance if overused. Right-angle adapters and remote heads solve packaging constraints in tight machines. Whatever you add, re-check image circle, distortion and µm/pixel, accessories change the conjugates you sized on paper.

Lock rings, set screws and threadlocker after the golden recipe is proven. A lens that “walks” open on a vibrating filler looks like a software regression every Monday.

How industries use industrial lenses

Packaging and FMCG

Moderate FOVs, mid-range C-mount lenses, frequent strobe lighting. Focus locks matter because operators change recipes across SKUs. Distortion rarely dominates; speed and contrast do. When date-code OCR shares the FOV with a seal check, size optics for the smaller text, not only the seal width.

Automotive and metal

Gauging stations lean toward low-distortion or telecentric optics; DPM and surface checks may use standard fixed lenses with carefully aimed lights. Oil, vibration and long cable runs favour rugged mounts. When a robot presents parts at varying heights, depth of field and autofocus strategies compete with “just stop the robot longer.”

Electronics

High magnification, short WD, high-resolution sensors and demanding MTF. Macro and high-resolution C-mount or specialized optics are common. Heat and ESD practices affect housing choice; the lens still has to clear nozzles and feeders. Pair optics early with the camera under Cameras.

Pharma secondary packaging

Reproducible FOV and locked focus support validation. Aggregation and code reading share cells with presence checks; document the optical recipe so field service can restore it after a crash without reinventing magnification.

Cost, quality and what “good enough” means

Teams sometimes under-buy the lens to protect the camera budget, then spend months fighting soft edges and unstable gauging. The inverse also happens: an expensive telecentric optic on a presence job that only needed a solid fixed C-mount and a better ring light. Match the optic class to the decision, detect, identify, or measure.

Practical budgeting allocates money to the weakest link in the chain: if samples are shiny, spend on lighting geometry and polarizers; if you measure to tens of micrometres, spend on telecentricity and mechanical stability; if you only need label presence, a well-chosen fixed lens plus strobe usually outperforms a high-megapixel camera with a soft kit optic.

Define acceptance with golden, borderline and dirty samples under the proposed lighting. If a human cannot see the feature crisply in the live image, software will struggle too, regardless of the SKU name on the lens barrel. More category depth lives under Knowledge. When samples are ambiguous, run a short feasibility pass before freezing the optical BOM.

Integrating optics into the machine

Lens selection is not finished when the purchase order ships. Mechanical design must keep the optical axis stable: rigid brackets, vibration isolation where needed, cable strain relief that does not torque the camera body, and access for cleaning windows without losing focus. Document WD, focal length, iris setting and filter stack in the machine’s vision recipe book.

Trigger timing, strobe pulse width and exposure must fit the depth of field and motion blur budget you accepted when choosing aperture. Multi-camera cells need consistent optical recipes across lanes so MES comparisons mean something. When architecture is still open, onboard smart tools versus PC libraries, settle that with smart vs PC vision before freezing mounts.

For discrete 2D stations the default camera family remains area scan; for ID-heavy cells, compare a dedicated reader against camera-plus-lens as described in the code-reader knowledge page.

Common first-system mistakes

• Buying megapixels first, then discovering the kit lens cannot resolve them.
• Ignoring image circle when swapping to a larger sensor.
• Setting FOV to the part only, no margin for placement error.
• Leaving focus and iris unlocked on a vibrating machine.
• Expecting software to fix perspective and distortion on a metrology job.
• Forgetting that lighting geometry and lens aperture share one photon budget.

Architecture still matters: a smart camera with integrated optics differs from a C-mount area scan plus separate lens. See smart vs PC-based vision when that choice is open.

Short glossary

FOV

Field of view, world width/height imaged onto the sensor.

WD

Working distance, lens-to-object distance (definition varies by vendor).

Image circle

Diameter of usable image projected by the lens.

C-mount

Common industrial lens mount and flange standard.

Telecentric

Optics with near-constant magnification and reduced perspective error.

MTF

Modulation transfer function, how well the lens transfers contrast at a spatial frequency.

f-number

Relative aperture, light gathering vs depth of field and diffraction.

FAQ: Computer Vision lenses

Next steps

You now have the optics model: FOV and WD set focal length with the sensor; image circle and mount must match; distortion and telecentricity matter when you measure; resolution class must keep up with the camera, and lighting finishes the optical path.

Related: Knowledge, area scan, feasibility.