What Camera Resolution and Lens a Factory Floor Needs

For most factory floors, a 4MP camera with a varifocal 2.8–12mm lens is the practical default. Match resolution to the pixels-per-metre your task needs — roughly 250 px/m to identify a face, 125 to check a helmet, 25 to count bodies — then pick the lens to place that scene, and stop. Megapixels rarely fix what light, shutter speed and dust break.

More megapixels feel like the safe buy, so plants routinely overspend on 8MP cameras that come back grainy and dark at night and chew storage — when a well-specified 4MP camera and the right lens would have done the job for half the money. The number that decides everything is not megapixels or lens millimetres in isolation. It is pixels per metre (px/m) on the thing you need to see — a helmet, a face, a forklift lane, a machine's status lamp. Get that number right and the spec sheet stops mattering.

Start with the task, not the spec sheet

The security industry has a published standard for "how much detail is enough": IEC 62676-4, known as DORI — Detection, Observation, Recognition, Identification. Each level maps to a minimum pixel density measured across one metre at the target's distance (Axis pixel-density white paper, IEC 62676-4):

The goalposts just moved. The revised standard, IEC/EN 62676-4:2025 (published 9 Oct 2025), adds a parallel ladder called OODPCVS and raises the bar for reliable, automatic face or number-plate identification from 250 px/m to a new "Validate" level of 500 px/m — precisely because 250 falls apart in low light and on moving subjects (Axis: from DORI to visual performance; Hikvision on OODPCVS). The old 250 DORI number is unchanged and still valid for a human reading a still image — but if you are buying face-ID positions against 250 alone in 2026, plan for closer to 500 px/m where you want software-grade certainty.

Most factory-floor value lives lower down, in the Detection–Recognition band — counting output, spotting a stopped cell, checking PPE. That is exactly why 4MP is usually enough and 8MP is usually waste.

2MP vs 4MP vs 8MP: what resolution buys you (scene width)

Resolution is horizontal pixels to spread across your scene. The wider one camera covers, the fewer pixels land on each metre — so resolution and coverage trade off directly. Scene width = horizontal pixels ÷ the px/m your task needs. From that formula:

Scene-width limits set by pixel count alone, computed from the IEC 62676-4 px/m thresholds; the lens then decides how far off that scene sits.

4MP is where the money stops working harder. The extra pixels over 2MP let one camera cover a wider bay while still reading helmets and faces — the default we recommend for a mixed floor. 2MP still earns its place for short-range overview, aisle watch and gate detection. 8MP earns its cost only when one camera must cover a very wide area and hold Identification density — a long dispatch dock on a single camera. Its hidden costs: bigger sensors that go noisy in low light unless well specified, and materially more storage and bandwidth (it scales roughly with pixel count at a given codec and bitrate). On a dim shop floor an 8MP camera often looks worse at night than a good 4MP one — here's why.

The trap the spec sheet hides: motion, light and shutter

DORI px/m assumes a static, well-lit target. Nobody on a factory floor is static. Under a dim 65-lux bay, auto-exposure drops the shutter to around 1/30s to gather light; a walking worker or a passing forklift then smears across many pixels. A camera that clears 250 px/m Identification on a test chart delivers only Recognition-grade blur on anything moving — effective resolution on a live floor is capped by shutter speed and light, not megapixels. For moving people and vehicles (PPE checks at a walkway, forklift lanes), specify a minimum shutter of 1/100–1/250s. That forces more light and a faster lens, which is the whole argument in one line.

Two camera-side specs decide dark-bay performance, and every low-light guide names them — so put them in your RFQ:

• Lens aperture / f-number. An F1.0–F1.6 lens gathers 2–4× the light of an F2.0. The catch nobody flags: varifocal lenses are usually slower and lose light at the tele (zoomed-in) end, so "buy varifocal everywhere" carries a hidden low-light cost — in your darkest bays a fixed fast lens can beat it.
• Sensor size / pixel pitch (µm). This is the mechanism behind "8MP looks worse at night": same sensor + 4× the pixels ≈ a quarter of the light per pixel ≈ more noise. Read pixel pitch and the minimum-illumination (lux) rating, not just the megapixel count. Where daylight windows meet dark bays, specify WDR.

The factory tax the spec sheet never mentions

The office-camera guides stop at resolution and lens. A factory doesn't. Before you finalise a spec, price in the environment:

• IP66 against dust and washdown; sealed housings for cement, flour and foundry particulate that films a dome lens within weeks.
• IK08–IK10 impact rating near forklift lanes and foot traffic.
• Ambient-temp rating — furnace bays and metal roofs run past 50°C, where consumer cameras cook.
• Vibration-rated mounts near presses, compressors and conveyors. And here is where blanket varifocal advice breaks: the varifocal you tuned perfectly drifts out of focus after a month bolted next to a stamping press, because vibration walks the zoom/focus set-screws. Spec vibration-damped mounts there, or use a fixed lens and accept the trade.

The lens, not the megapixels, sets your field of view

Focal length (mm) sets the horizontal angle of view — but so does sensor format, and the two interact. The angles below assume a typical 1/2.8" sensor; a smaller 1/3" or 1/4" sensor narrows every figure, which is one more reason varifocal removes the guess (CCTV lens comparison, CCTV Camera Pros):

Wide lens = wide coverage but low detail at distance; long lens = high detail but a narrow slice. A 2.8mm camera aimed across a large hall detects motion but won't identify a face at the far wall — pixels spread too thin. A 6mm reads a helmet at 12 m but sees only a corridor.

This is why a varifocal / motorised 2.8–12mm lens is the pragmatic default for most positions: on site the mount is 30 cm off the drawing and a stacked pallet blocks the shot, and the zoom ring fixes it instead of a re-order. It costs more than a fixed lens but saves a second truck-roll — just remember the two exceptions above: your darkest bay and your highest-vibration mount.

One machine-monitoring note for Indian mains

If a camera is watching an indicator lamp or rotating equipment (spindles, fans, belts), enable anti-flicker / 50 Hz mode. On a rolling-shutter CMOS sensor under India's 50 Hz mains with LED or fluorescent lighting, you get banding, and a blinking status LED can alias to "off" at the wrong shutter phase; rolling shutter also skews fast rotating parts. It is a small setting that decides whether "read the machine's status lamp" actually works.

Indicative Indian prices (as of July 2026)

Treat these as a planning band, not a quote — pricing varies by dealer, batch and night-vision tier (CCTV price guide, TheNcix):

On live same-brand listings the step from 2MP to 4MP is often only a few hundred to ~₹2,000 per camera — small enough that 4MP is the right default anywhere you might later want PPE or face detail. Budget separately for the essentials: a PoE switch on a UPS, H.265 / H.265+ encoding (roughly halves storage and bandwidth vs H.264 for the same image), and your mount (turret/dome for ceilings, bullet for walls). On a retrofit, the real ceiling is often not disk — it's piling 8MP streams onto existing switch uplinks, NVR ports and cabling.

Compliance note: cameras newly sold or installed in India from 1 April 2026 must meet BIS/STQC ER-01 security requirements (BIS CRS CCTV guidelines); specify ONVIF Profile T for interoperability with third-party recorders and analytics (ONVIF Profiles). Hikvision/Dahua face procurement restrictions on government and critical-infrastructure sites and heightened BIS scrutiny — use them here only as a price reference; private-factory eligibility differs, so confirm before you standardise on a brand.

Don't forget light — resolution can't fix a dark shop

Pixels are useless without photons. Walk the floor at 8 pm, not on the vendor's noon demo: plenty of bays run on one tired tube-light and a skylight that does nothing after dark — that's exactly where an 8MP camera turns to grain. Beyond the image, adequate lighting is a statutory duty: the Factories Act, 1948, Section 17 requires sufficient and suitable lighting wherever workers work or pass (India Code), with the Model Rules under §17(4) prescribing general illumination of not less than 65 lux over regularly-worked interior areas (Model Rules, DGFASLI). The exact lux duty is set by each state's factory rules; 65 lux is the model figure. Spend on lighting and lens quality before spending on raw megapixels.

How to decide, position by position

You don't buy one spec for the whole plant — you buy per position, per task:

1. Name the task and its px/m. Count = 25 (Detection). PPE = 125 (Recognition). Face/label = 250, or 500 for software-grade face-ID (Identification / Validate).
2. Measure the scene width that camera must cover.
3. Pick resolution so pixels ÷ scene width clears the px/m — usually 4MP.
4. Pick the lens to place that scene at the right distance — varifocal unless it's your darkest or most vibration-heavy spot.
5. Check motion and light at the darkest shift: enough shutter (1/100s+) and lux to keep the target sharp, plus IP/IK/heat/vibration for the position.

Worked example. A 20 m assembly bay you want to recognise people across (helmet on/off, 125 px/m): 2560 px ÷ 20 m = 128 px/m — a single 4MP camera clears it. Want to identify faces across the same 20 m? 2560 ÷ 20 = 128 px/m, well short of 250 — so you either drop to a ~10 m scene per camera (two cameras), or step up resolution. That is the recurring over-buy we see: 8MP thrown at a bay where a 4MP-plus-varifocal, placed right, cleared Recognition for half the cost.

Doing this by hand across 20–60 positions is where projects stall or over-buy. That "which camera, which lens, for which task, where" mapping is the gap Mama closes: record a short phone walkthrough, and it reads the floor — bays, sightlines, hazards, where the machines actually sit — and hands back two things: a floor plan, and a camera-per-position spec with the resolution and lens each task needs. The layout and the spec, in a day, no site survey.

Bottom line

• 4MP + varifocal 2.8–12mm is the default for most positions; 2MP for short-range overview, 8MP only for wide-area-and-detail.
• Buy per position against px/m — 25 count, 125 PPE, 250 face, 500 for software face-ID.
• Light and shutter beat megapixels — spec 1/100–1/250s and F1.0–F1.6 for moving/dark scenes; check pixel pitch and lux, not just MP.
• Pay the factory tax — IP66, IK08–IK10, >50°C rating, vibration-damped mounts (which is why blanket varifocal is wrong at press mounts).
• India rules — 65 lux (Factories Act §17), BIS/STQC + ONVIF Profile T from 1 Apr 2026.

FAQ

Is 4MP always better than 2MP for a factory? Not always, but it's usually the right default. 2MP is fine for short-range overview, counting and detection. On live same-brand listings 4MP costs only a few hundred to about ₹2,000 more per camera and gives headroom to identify people or read a machine indicator over a wider bay. Buy 2MP only where the task never needs more than overview.

Do I need 8MP / 4K cameras? Rarely. 8MP earns its cost only when one camera must cover a very wide area and still hit Identification density — like a long dock. Otherwise it costs more, needs more storage and bandwidth, and often performs worse in the low light of a shop floor than a well-specified 4MP camera. Same sensor plus 4× the pixels means about a quarter of the light per pixel, so more noise.

2.8mm or 6mm lens — which should I pick? 2.8mm gives a wide ~90–110° view for overview and small areas up close; 6mm gives a narrower ~55–65° view that reaches a specific line or machine 8–15 m away with detail (angles assume a ~1/2.8" sensor; a smaller sensor narrows them). Unsure? Buy a varifocal 2.8–12mm and tune it on site — except in your darkest bay, where a fixed fast (low f-number) lens keeps more light.

How do I know if a camera can identify a face at a given spot? Use IEC 62676-4: about 250 px/m for human Identification, 125 for Recognition (helmet on/off), and 500 px/m ("Validate", added in the 2025 revision) for reliable automatic face recognition. Divide the camera's horizontal pixels by your scene width in metres; if it clears the threshold for your task, you're covered — provided the shutter is fast enough that the subject isn't blurred.

What else should I budget for besides the camera? A PoE switch on a UPS, cabling, H.265 encoding to cut storage, the right mount (turret/dome for ceilings, bullet for walls), IP66/IK-rated housings for dust and impact, and adequate lighting — Section 17 of the Factories Act, 1948 requires it. Specify BIS/STQC-compliant, ONVIF Profile T cameras so the system is legal and interoperable.