Analog vs digital cameras for machine vision: how to choose

Four cameras on a washdown line, an aging frame-grabber cabinet, and a quote that still lists analog options — that is where the analog-versus-digital question actually gets decided. The decision is simpler than the discussion around it: digital GigE Vision and USB3 Vision cameras win every new build on resolution, cabling and integration effort, and analog survives only as an addition to a working frame-grabber system. Below you get the pixel arithmetic that sizes the camera, the cabling and trigger differences that size the installation, and one decision per case you can put straight onto an order.

Are analog machine vision cameras still worth buying new?

No: for a new build, digital wins on every spec line — a current GigE Vision or USB3 Vision camera carries 2.5 to 25 MP over one standard cable, where an analog composite feed delivers about 0.31 MP (640 × 480) through a frame grabber. That pixel gap, 8× at the small end and 80× at the top of the range, settles most purchases before any other argument is heard.

The analog ceiling comes from television history, not from the sensor. RS-170 (the EIA 525-line standard) and CCIR (625 lines) fixed the frame timing and the luminance bandwidth decades ago, so a frame grabber samples the composite waveform into about 640 × 480 pixels on RS-170 and about 768 × 576 on CCIR, whatever the sensor in front of it can resolve. A digital camera removes the ceiling: the sensor defines the pixels and the interface moves them losslessly, which is why resolution climbs from under 3 MP to over 25 MP across today's area scan cameras.

Digitizing at the sensor also moves the noise argument. Sensor quality, not grabber timing, decides low-light performance, and makers publish it as EMVA 1288 measurements you can compare directly — this walkthrough of low-light camera selection from EMVA 1288 data shows the comparison.

A 20 MP mono camera such as the MV-CS200-10UM pairs a 1" IMX183 sensor and C-mount with a USB3.0 link.

Twenty megapixels is about 64 analog frames of data: 20 MP ÷ 0.31 MP ≈ 64. If the station only flags presence, most of that margin is wasted — the real question is never 'more pixels' but 'enough pixels', and the pixel budget below sizes it. If the whole line is new rather than an extension of an analog cabinet, start from what a machine vision system is and how to choose one and place the camera choice inside it.

What is actually different between analog and digital cameras?

The digitization point is the difference: an analog camera sends a composite video waveform down coax and the frame grabber samples it into roughly 640 × 480 pixels, while a digital camera converts at the sensor and ships finished pixels over GigE, USB3, CoaXPress or Camera Link. Everything else in the comparison follows from that one distinction.

Two consequences follow from the table. The analog path concentrates cost and failure in one card: four cameras share the grabber's timing, so one mis-set input standard degrades several channels at once. The digital path moves control onto the data cable — exposure, gain, ROI and trigger mode are SDK fields under GigE Vision and USB3 Vision, not back-panel switches. And when a camera outgrows one GigE port, the digital ladder continues through 10GigE, CoaXPress and Camera Link on frame grabbers; composite video has no rung above it.

Which one fits an existing analog frame-grabber system?

Stay analog only while the installed grabber has spare channels — many analog cards take four cameras per board — and the images still pass inspection; plan the digital migration now, because analog capture hardware is out of production. Count the spare channels first, and hold every addition to the card's input standard.

The conditions are strict. The camera's scanning standard must match the card's input — RS-170 versus CCIR, interlaced timing — and analog cameras now come only from legacy suppliers, so two cameras of the same type from different production batches can sample slightly differently. Failure modes, not age, tell you when the analog path is finished: a frame split in two by a black line is cable and sync, and it has its own diagnostic walk-through; rolling bands across the picture point at grabber settings, covered in this scanline distortion check. When the questions move from cameras to card configuration, analog frame grabber integration with vision software is the companion page.

The compatibility checks are listed in camera frame grabber compatibility for resolution upgrades.

How much resolution does your inspection actually need?

Work it as a pixels-per-feature budget: most inspection tasks need 3 pixels across the smallest feature they must flag, so a 0.3 mm defect sets 0.1 mm/pixel and a 120 mm field demands 1200 × 900 pixels — 1.08 MP — before any interface question. Detection lives at 3 px per feature; measurement wants 5–10 px so edge algorithms interpolate on real structure.

Worked example: a 120 mm × 90 mm field, a 0.3 mm crack to catch. The budget: 0.3 / 3 = 0.1 mm/px. Pixels across: 120 / 0.1 = 1200; pixels down: 90 / 0.1 = 900; area check 1200 × 900 = 1,080,000 px ≈ 1.08 MP. An analog camera fails twice over. Coverage: 120 / 640 = 0.19 mm/px, so the crack spans 0.3 / 0.19 ≈ 1.6 px, under the 3 px floor. Tile size at 0.1 mm/px: 640 × 0.1 = 64 mm by 480 × 0.1 = 48 mm, which is 10,800 mm² ÷ (64 × 48 = 3,072 mm²) ≈ 3.5 tiles — four analog cameras for one station.

The LS-CMR-0166 resolves 12 MP (4000 × 3000) at 30 fps over USB3, on a 1/1.7" rolling-shutter IMX226.

Over the same field: 120 / 4000 = 0.03 mm/px, so the crack spans 0.3 / 0.03 = 10 px — about 3× the floor, enough to measure the crack instead of merely flagging it. One camera replaces the four analog tiles.

Size your own station with the calculator below; its result links straight into the camera finder filtered to sensors with enough pixels.

What changes on the cable and trigger side?

Cabling gets simpler and longer: GigE runs 100 m per cable segment with power delivered over the same wire by PoE, while analog coax needs one run per camera plus grabber-side trigger wiring and shows cable faults as noise bands or split frames. One connector per camera replaces the coax-plus-power-plus-trigger bundle.

The MV-GE2002P dual-port PoE NIC feeds data and camera power to two cameras on two patch leads.

Port count then scales with a capture card instead of a second technology. The MV-GE1104 is a 4-port GigE frame grabber. Four cameras land on one card, and the four ports must be told apart at commissioning — the swapped-channel trap covered in assigning cameras to ports reliably in a multi-camera system. Two cameras can also share one host port with a bandwidth budget, as worked in two GigE cameras on one host.

Budget the port itself. A 12 MP frame is 12 MB at 8-bit mono, so the 30 fps example pulls 12 × 30 = 360 MB/s. One GigE port delivers about 118 MB/s in practice (1000 Mbit/s ÷ 8 = 125 MB/s raw, less after overhead), which is why 12 MP parts ship on USB3 or 10GigE instead. Host driver settings decide how much of the link you actually get; the measured effect is quantified in GigE Vision high performance NIC driver: how much you gain.

Triggering moves from the card to the camera. Analog lines wire the trigger into the grabber, which paces every channel together; a digital camera carries its own trigger input and times each exposure itself — better for staggered stations, and a trap during conversion because the factory default is free run. Set the trigger deliberately per camera (hardware trigger configuration per camera), and match the electrical level and cable when a signal generator drives it (camera trigger cable selection for a signal generator).

Does a washdown line force an IP67 camera?

It is a separate decision from analog versus digital: most integrators keep a standard camera and mount it in a washdown-protective housing, because IP67 (30 minutes at 1 m immersion) and jet-wash ratings test different things per IEC 60529. The camera question and the enclosure question stay independent.

The rating language matters more than the number. IP67 means dust-tight plus 30 minutes at 1 m of immersion; jet and wash-down protection live under IP66 and IP69K, which are different test rigs. The cabinet world has the same mismatch between NEMA and IP codes, and the NEMA 4X versus IP66 ratings comparison untangles it for housings and enclosures. A washdown housing holds a standard digital camera in service with its matched glands and connectors, so the camera fleet stays uniform and the housing, not the camera, carries the wash rating. Treat camera plus housing as one rated assembly: the rating holds only with the cable entries the housing lists for your wash class.

When is analog still the right call?

One case only: a live analog frame-grabber line with spare channels and passing images, where swapping to digital buys nothing this quarter — and even then you order analog cameras as a stopgap with a written migration date, not as new capability. The stopgap has rules of its own.

Buy against the card's input standard, and write the migration date into the purchase: analog capture boards are no longer made, so spare-card inventory, not camera supply, becomes the binding constraint within a couple of procurement cycles. Then run the conversion in this order:

1. Inventory the cabinet: channels per card, input standard per channel (RS-170 or CCIR), trigger wiring.
2. Re-run the pixel budget per station; note where 640 × 480 passes with no margin, because those are the stations a digital camera upgrades immediately.
3. Convert one station first: mount the digital camera beside the analog one on the same field, run both for a week, and compare defect calls rather than pictures.
4. Fix the camera-to-port map before cutover day so channels cannot swap.
5. Move channel by channel, keeping the trigger source, and retire the analog card when its last channel converts.

A full camera-for-camera replacement is walked through in replacing an obsolete analog NTSC frame grabber with USB3 Vision; where the analog feed ends in an embedded controller instead of a PC card, weigh decode-and-keep against replacement in analog camera on an embedded vision controller: decode or upgrade.

Which camera do you order for each case?

New or rebuilt line: order digital sized by the pixel budget — 2.5 MP clears the 1200 × 900 example with margin, 12 MP when one camera must cover several stations. Extending a live analog line: match channels and timing now, migrate channel-by-channel later. The full map:

For the first row, the MV-CU020-90GC is a 2.5 MP global-shutter color GigE camera on a 1/1.7" GMAX4002 sensor.

Against the example budget: 2.5 MP ÷ 1.08 MP ≈ 2.3, so a 1200 × 900 layout fits with margin on every side. If color adds nothing, a mono sensor of the same class usually buys more pixels — the sensor trade-offs are compared in USB3 machine vision camera selection: compare sensor trade-offs.

Check before you order:

• Pixel budget: at least 3 px per feature to detect, 5–10 px to measure.
• Bandwidth: camera MB/s against the port's practical rate, 118 MB/s per GigE port.
• Triggers: one input per camera, electrical level matched to the source.
• Ports: write the camera-to-port map before commissioning, not during it.
• Washdown: housing and its listed cable entries rated as one assembly.

Take the pixel budget and the station list to the camera configurator, search cameras by interface and resolution, or browse frame grabbers by port count. If a mixed line has to keep running through the changeover, the contact page is the route to a layout review. The cameras topic page collects the camera-selection guides.