Vision sensor vs vision system: choosing by inspection complexity

Every inspection station starts at the same fork: a self-contained vision sensor that answers one question, or a component vision system you program and light yourself. Most wrong purchases happen exactly here — sensors bought for gauging jobs they cannot do, and PC-based systems built for a presence check a sensor would settle in ten minutes. This comparison gives you the one test that decides, the pixel arithmetic behind it, a worked example on a real bracket, and a verdict for the common cases. By the end you will know which side of the fork your line is on and what to order for it.

The decision in one question

Write down the sentence your process consumes. If it is "bracket present: yes" — a bit on a PLC input — buy a vision sensor. If it is "hole diameter 24.03 mm", a code string, or an image kept for traceability, buy a vision system. Everything else follows from that fork: setup time, optics, resolution, wiring and who maintains it.

The asymmetry settles borderline cases. A full system can always reduce its numbers to a pass/fail bit, so it covers a sensor's job. A sensor cannot emit a number worth acting on, because nobody chose its fixed lens and pixel count for gauging. So the rule is:

• The output is one bit, today and for the life of the station: vision sensor.
• The output is a dimension, a position in millimetres, a code string or a stored image: vision system.
• One bit today, numbers likely next quarter — traceability, SPC, a customer claim: vision system, or a networked smart camera if no PC may live at the station.

When both genuinely work, the sensor wins on integration time and swap time. When the process may ever need the number, the system wins on day one.

What you are choosing between

A vision sensor is one housing: lens, illumination, imager, processor and I/O. You aim it at the check, teach a reference, and its discrete outputs drive a PLC. The vision sensor range runs from 0.3 MP fixed-focus units to 12 MP models and beyond, most on integrated M12-class lenses over Fast Ethernet or GigE. The MV-SCB007EC-12S-WBN is a 1.6 MP color vision sensor with a 12 mm lens and manual focus. The MV-SCA002M-03S-WBN-SR is a 0.3 MP mono unit with a 3 mm lens and a 240 mm focus distance. Both are presence-class tools: teach, wire, run.

A smart camera sits between the two camps: self-contained like a sensor, but programmable with multiple tools and a network link. The MV-SC5060XM-08M-WBN is a 6 MP mono smart camera with an 8 mm lens on GigE, and the LS-SMC-0063 adds a built-in white light at 60 fps. Use one when several checks must share a single image but no controller may sit at the station; the tool-set question — rule-based or AI — is covered in AI vs rule-based smart camera inspection.

A vision system is components you choose: an area scan camera, a lens picked for the field of view, your lighting, and a vision controller running the tools and storing the images. The MV-CH120-11UC is a 12 MP global-shutter camera with a 1.1-inch sensor format and a C-mount. The MV-VC3301H-128G66 is a vision controller with an Intel Core i3 processor, 8 GB of memory and four built-in light channels — strobe timing and processing land in one box.

Vision sensor vs vision system at a glance

The numbers that decide it

Both sides take images; the difference is how many pixels each can spend on the feature that matters. Two rules cover almost every case:

• Presence: the smallest feature you judge must span at least 4 to 5 pixels. Below that, lighting wobble and vibration turn the check random.
• Measurement: the tolerance band itself must span at least 5 pixels, so sub-pixel edge fitting (good for about 0.1 to 0.2 px of repeatability) has something to resolve.

Counting pixels on target is three lines of arithmetic:

Sensor width is the active width in millimetres and WD the lens-to-part working distance. The lens focal length calculator and the camera resolution calculator run the same sums with your numbers; for matching lens resolution to a feature size, see lens selection based on resolution.

Worked example: one bracket, two questions

The part: a steel bracket with a critical width of 24.00 ± 0.10 mm, sitting anywhere within ±0.5 mm of nominal height, passing the camera at 300 mm/s. Two different questions get asked about it, and they get different hardware.

Question 1: is a bracket here?

Take a 0.3 MP presence-class unit (640 × 480 pixels) with a 3 mm lens focused at 240 mm. A 1/6-inch sensor is about 2.4 mm across, so:

• Magnification m = 3 / (240 − 3) = 3 / 237 = 0.0127
• Field of view = 2.4 mm / 0.0127 ≈ 190 mm
• Pixel on target = 190 mm / 640 px ≈ 0.30 mm

The 24 mm bracket spans about 80 px — a solid presence signal. The ±0.10 mm tolerance spans 0.33 px. Nothing in that image measures anything; the answer is a bit, and the bit is reliable. Verdict: vision sensor.

Question 2: is the width 24.00 ± 0.10 mm?

Now take a 12 MP camera, 4096 × 3000 pixels at a 3.45 µm pitch, so the sensor is 14.1 mm wide. Allow a 60 mm field of view around the 24 mm part:

• m = 14.1 / 60 = 0.235
• Pixel on target = 3.45 µm / 0.235 = 14.7 µm ≈ 0.015 mm, so the full 0.20 mm tolerance band spans about 14 px — nearly three times the 5 px rule
• Focal length at 200 mm WD: f = 200 × 0.235 / 1.235 = 38 mm, so fit a 35 mm lens; the working distance becomes WD = 35 × 1.235 / 0.235 ≈ 184 mm
• Depth of field for the ±0.5 mm height error: DoF ≈ 2 × f-number × pixel pitch × (1 + m) / m² = 2 × 8 × 0.00345 × 1.235 / 0.0552 ≈ 1.2 mm at f/8 — enough
• Motion at 300 mm/s: blur of one pixel (14.7 µm) caps exposure at 14.7 µm / 300 mm/s ≈ 49 µs, so a 50 µs strobe pulse freezes the part

Six numbers, one decision: 35 mm lens, 184 mm working distance, 60 mm field of view, 14.7 µm per pixel, 50 µs strobe, f/8. The shutter matters too — a rolling shutter would smear a part moving at 300 mm/s line by line, so the camera is global shutter; global vs rolling shutter for moving targets covers the failure modes. For a full measurement cycle on a small part, see measuring points on a 5 mm part in under one second.

Which one for your line

• Presence, absence, counting, color match at one station, wired to a PLC input: vision sensor. Machine vision sensor selection for high-speed inspection covers the speed side, and this label inspection sensor selection guide walks a conveyor case.
• Any tolerance quoted in millimetres: vision system. That includes go/no-go gauging — the sensor route above was 15 times short of the 5 px rule.
• Several independent checks on the same part: one smart camera beats a row of sensors once you pass three checks, because each sensor needs its own mounting, wiring and reference.
• Code reading with data to keep: smart camera or system with Ethernet; the output side is its own job — see parsing tab-delimited results from a smart camera.
• Robot guidance and 3D profiling: system.
• Screw, turned-part and similar dimensional QC: system — this vision inspection selection for screw manufacturing guide covers the case end to end.

On wiring, a sensor's discrete outputs land straight on PLC inputs; smart camera to PLC over TCP shows the trigger and result handshake. A system sends results the same direction but richer — vision system to PLC makes the case for sending the verdict, not the image, inside the cycle.

Checks before you order

1. Write the output sentence first ("PLC input 3 energises" vs "width in mm to the database"). If you cannot write it, the choice is not ready.
2. Compute pixels on target with the formulas above; test the 5 px rule against your tolerance and the 4 px rule against your smallest feature.
3. Confirm a lens exists for your field of view and working distance, and that mount and sensor format match it.
4. For moving parts, derive the strobe length from the one-pixel rule and insist on a global shutter.
5. Check lighting margin at your f-number: closing from f/4 to f/8 costs two stops, which the 50 µs strobe pulse has to make up.
6. Decide who re-teaches the station when the line is down at night — technician for a sensor, vision engineer for a system — and stock the spare accordingly.
7. If a smart camera is in play, run the software lock-in check before the purchase order.

For the system route, machine vision lens and lighting selection: a test plan and stable 2D vision inspection systems turn these checks into a bench procedure.

When the usual answer is wrong

• "It's only presence" — until the smallest feature drops under about 0.5% of the field of view. On a 640-pixel sensor that is 3 px, below the presence rule. Then you need a higher-resolution sensor model or a system with chosen optics, even though the output stays a single bit.
• "It's only a limit check" — with a generous band it is a sensor job. A ±5 mm fill-level check on a 190 mm field of view spans about 17 px on the presence-class unit. The cutoff is the 5 px rule, not the word "measurement".
• Eight sensors on eight stations — one multi-point system wins the moment the plant wants line-level data or a single change-management path. Count the sensors before you count the pixels.
• A vision system bought for one stable yes/no at speed: slower to integrate, more failure points, and a sensor's deterministic discrete output is the better fit.
• A sensor with no image storage, bought into a regulated line: the audit will want the picture. Networked smart cameras close that gap without a full system.

Run the numbers above for your own part, then take the choice through the system configurator. If the fork is still unclear, send the output sentence and a part drawing through the contact page and we will size both routes with you.