Consumer camera vs machine vision camera: what you give up

Every integrator meets the question sooner or later: the lab already owns a 24 MP digital still camera, so why buy an inspection camera? The honest answer is that the two devices solve different problems. A consumer camera makes one good-looking frame of a scene a person chose. A machine vision camera delivers a known frame of a scene a machine controls, on a schedule measured in microseconds. This comparison covers the differences that decide automated inspection — triggering, exposure, drivers, timing — and names the cases where each camera is the right purchase.

The decision, up front

For any inspection that runs without a person pressing the button, buy the machine vision camera. A consumer still camera resolves plenty of megapixels, but it gives you no deterministic trigger, no guaranteed frame rate over its tether, exposure you cannot set to a number, and no driver path your vision software can enumerate. Those four gaps are structural, not fixable in software, and they are why a 2.5 MP industrial area scan camera beats a 24 MP consumer body on a production line. Keep the consumer camera for static, human-paced work: documentation photos, photogrammetry of a stationary object, one-off lab captures. Anything that moves, repeats or answers to a PLC needs the industrial camera.

Side by side: what you give up

The pattern in every row is the same: the consumer camera hides its settings to protect image quality, while the industrial camera publishes them so you own the pixel values.

Drivers: file transfer is not image acquisition

The driver question is where consumer-camera projects stall. Most consumer bodies expose pictures through PTP (Picture Transfer Protocol), the file-transfer protocol operating systems use for cameras and phones. A few makers publish tethering SDKs, and third-party libraries wrap PTP for measurement software. What arrives is a finished photograph — a few frames per second, delivered after focus and metering complete. You cannot schedule the frame, and you often cannot switch off the processing that made it pretty.

An industrial camera inverts that relationship. GigE Vision and USB3 Vision define discovery, streaming and control on the wire, and GenICam publishes every feature — exposure time, gain, trigger source, region of interest — as a standard feature tree your software reads at connect. Any compliant package controls the camera without maker-specific code; what GigE Vision guarantees for software compatibility covers that contract in detail. Interface choice then becomes a bandwidth and cabling calculation rather than a driver hunt, and the GigE Vision camera selection guide walks that arithmetic with numbers.

Triggering: the part decides when

A production camera takes its cue from the process. An industrial camera has a trigger line — usually opto-isolated — wired to a photoelectric sensor, an encoder or a PLC output, and the delay from trigger pulse to exposure start is specified in microseconds. Hardware triggering with external pulses covers the timing chain, and software trigger ownership settles who commands the frame when the PLC and the PC both can.

A consumer body has nothing to wire to. A remote release fires one still capture with tens of milliseconds of latency and no schedule, and a tethered still takes seconds per frame once focus, metering and file write complete. On a conveyor at 200 mm/s the part travels 200 to 400 mm during that cycle — it has left a 120 mm field of view long before the file lands.

Exposure control: what auto mode costs

Auto exposure on a consumer camera meters the scene and picks shutter, aperture and ISO for every frame, then applies gain, gamma, sharpening and noise reduction before writing an 8-bit JPEG. For a photograph that is the point. For measurement it is three losses at once: exposure drifts frame to frame so grey values move, edge sharpening shifts sub-pixel feature locations, and JPEG rings along high-contrast edges — exactly the edges a metrology tool locates.

An industrial camera hands you the numbers: exposure set in microseconds, gain locked near zero, raw linear pixels out. The exposure, gain and balance settings guide gives the working procedure, and why video exposures differ from still images explains the frame-rate coupling auto mode hides. Because exposure is just a register, you can also change it per product on the fly — runtime exposure control without a driver reload shows the pattern.

Worked example: a label at 200 mm/s

Numbers make the gap concrete. The task: read a printed label 120 mm × 90 mm riding a conveyor at 200 mm/s, where the smallest printed stroke is 0.2 mm and the working rule is 3 pixels across the smallest feature.

• Pixel size: 0.2 mm / 3 = 0.067 mm per pixel.
• Pixels across: 120 / 0.067 = 1800 wide and 90 / 0.067 = 1350 high, so 1800 × 1350 = 2.4 MP. A 2.5 MP camera does the job; 24 MP adds nothing the task can use.
• Lens: a 1/1.7" sensor is about 7.6 mm wide, so magnification is 7.6 / 120 = 0.063. At a 300 mm working distance, f = 300 × 0.063 / 1.063 = 17.8 mm — an 18 mm C-mount lens (estimate; the exact sensor width moves it a millimetre or two).
• Exposure: allow 0.5 pixel of blur, which is 0.5 × 0.067 = 0.033 mm. At 200 mm/s, t_max = 0.033 / 200 = 0.00017 s, about 170 µs. Set 150 µs and drive the light as a strobe.

The MV-CU020-90GC is a 2.5 MP GigE camera with a 1/1.7" sensor, a C-mount and a global shutter — the exact envelope this example calls for. A consumer body faces the same arithmetic and fails it three ways: its fastest shutter, typically 1/4000 to 1/8000 s (250 down to 125 µs), is reachable only in full manual; auto mode will not choose it; and nothing tells the camera when the label is in the window.

Where the megapixels argument fails

The one real argument for a consumer body is sensor resolution for the money, and on a line it argues against itself. More pixels mean more data per frame, so frame rate falls: a 45 MP still body tethered delivers a few frames per second of JPEG, while industrial sensors stream continuously. The LS-CMR-0166 is a 12 MP USB3 camera that runs 30 fps at 4000 × 3000 pixels — far past what the label task needed, still real-time, still triggerable. The MV-CS200-10GC is a 20 MP GigE camera with a 1" sensor, a C-mount and a rolling shutter: consumer-class resolution that still enumerates, triggers and streams. When the part is a continuous web rather than discrete pieces, the answer moves to line scan — area scan cameras in line scan mode covers that transition.

Three industrial cameras at three resolution tiers make the point directly:

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

Less hardware, not more: smart cameras and the host card

If the PC is the part of the industrial design you were hoping to avoid, drop one level further instead of retreating to a consumer camera. Smart cameras and vision sensors run the inspection on the device and answer over Ethernet, with the same trigger discipline as a frame-grabbed system:

The MV-SC5060XM-08M-WBN is a 6 MP mono smart camera with an 8 mm lens and a GigE interface — sensor, optics and inspection logic in one housing. On the host side, the deterministic path ends in a PoE NIC or frame grabber rather than whatever chipset the office PC shipped with:

The MV-GE2002P is a dual-port PoE GigE network card — two camera ports, powered over the same cable. If your software sees the card but not the camera, why one viewer works and another fails is the first place to look. When the station grows past a few cameras, assigning cameras to ports reliably keeps discovery deterministic at startup. Browse the host interfaces in the frame grabber finder alongside the camera finder.

When the consumer camera is the right answer

Commit to the industrial camera for automation, and keep the consumer body for the three jobs it genuinely does better:

• Documentation and traceability photos of an assembly, taken by a person and filed with the record. Resolution and colour science matter; timing does not.
• Photogrammetry or texture capture of a stationary object, where 40-plus megapixels beat any industrial sensor and nothing moves between frames.
• Bench proof-of-concept before any automation exists — with the understanding that none of its timing conclusions transfer to production.

A USB webcam is the same decision one rung down: it clears the driver hurdle because it speaks UVC, but the fixed lens, rolling shutter and auto exposure remain, and choosing the right capture path is its own integration question.

Checks before you order

Work down this list against the data sheet; every item has ended a consumer-camera pilot at some point.

• Trigger input. A real hardware line with a stated trigger-to-exposure latency in microseconds. A remote-control app is not a trigger.
• Exposure range. Times settable in microseconds, short enough for your part speed, with gain you can lock.
• Shutter type. Global if anything moves; rolling only when the part is clamped during exposure.
• Output. Raw or standard pixel formats, with no JPEG anywhere in the chain.
• Interface. GigE, USB3, 10GigE or CoaXPress to match your cable runs and frame rate.
• Lens mount. C-mount or larger, so you can buy the focal length the field of view demands. For small fields the lens dominates the choice — selecting a compact machine vision camera for a 4 cm field works that case end to end.
• Life cycle. The camera should be orderable for years, with an end-of-life notice you can plan around.

If you would rather hand the arithmetic to a team that does it daily, scope the system first, then talk to us — expert support comes with every order.