How to Choose an Industrial Camera (An Engineer's Step-by-Step Guide)

Choosing an industrial camera is not a spec-sheet comparison — it is a chain of decisions where the cheapest mistake costs you weeks. This is the method we run on every Selection Sprint, in seven steps.

Step 1: Define the task before the camera

Write down, in one paragraph:

• What you inspect or measure (the part, the defect or the dimension).
• How big the field of view is and the working distance to the part.
• How fast the line moves (parts per minute, or metres per minute for web inspection).
• How accurate the result must be (detection only, classification, or metrology in microns).
• Where it lives (cleanroom, heat, vibration, washdown).

Everything downstream — resolution, shutter, interface, lens, lighting — falls out of this paragraph. Skip it and you will buy a camera that photographs the problem instead of solving it.

Step 2: Resolution — work backwards from the smallest feature

The rule of thumb we use:

• Detection (is it there? is it broken?): at least 3–5 pixels across the smallest feature you must see.
• Measurement / metrology: 10+ pixels across the smallest dimension you must measure.

So the minimum sensor resolution follows from:

required resolution = (field of view ÷ smallest feature) × pixels-per-feature, on each axis.

Two traps here. First, more resolution is not free: it costs frame rate, data bandwidth, lens quality and money — and a sharper sensor on a bad lens is a blurrier image. Second, “5 MP” means nothing without the field of view; the same 5 MP sensor is generous on a 50 mm view and inadequate on a 500 mm view.

Step 3: Frame rate and shutter — global vs rolling

• If the part moves during exposure, you need a global shutter. Rolling shutter skews moving parts (the classic “leaning building” effect).
• If the part is stationary during exposure (strobe-lit), rolling shutter is usually fine and cheaper.
• Compute the frame rate you actually need: line speed ÷ (pixels across motion direction) × margin. Add 30–50% margin for trigger jitter and future line speeds.

Step 4: Interface — match the data pipe to the camera

The interface decision is really a cabling and host decision: how far the camera sits from the PC, how many cameras share the host, and what your software stack speaks. All of these standards are vendor-neutral (GenICam), which is exactly why a multi-brand shortlist is possible.

Step 5: Colour or monochrome?

• Colour when the defect is defined by hue (printing, display Mura, wire colours, solder grading).
• Monochrome when the signal is contrast, shape or edges — mono sensors are 2–3× more light-sensitive at the same resolution and often cheaper.

If you need colour, budget for good, stable illumination: colour algorithms inherit every flicker and colour-temperature error in your lighting.

Step 6: Lens and lighting — where inspections are actually won

A mediocre camera with the right lens and light beats a great camera with the wrong ones. Every time.

• Lens: match resolution (the lens must resolve what the sensor sees), mount, working distance and field of view. Telecentric optics for metrology; standard C-mount for everything else.
• Lighting: dark field for scratches and dents, bright field for presence, backlight for silhouettes and dimensions, coaxial for flat surfaces, structured light for 3D. The wrong geometry hides the very defect you are hunting.

This is also where vendor-neutral advice pays: the “camera” decision is usually a camera + lens + light decision, and only someone who tests combinations will get it right.

Step 7: Verify before you buy

Even after all seven steps, one unknown remains: how this specific camera behaves with your optics, your light and your part. That is what a bench test is for — resolution with your lens, noise at your gain, colour under your light, trigger behaviour at your line speed. A few hundred dollars of testing is cheap insurance against a wrong order.

The short version

1. Define the task and constraints first.
2. Size resolution from the smallest feature, not the marketing number.
3. Match shutter and frame rate to the motion.
4. Choose the interface from cabling and host reality.
5. Colour only when the defect is colour.
6. Spend real effort on lens and lighting.
7. Test the final combination before committing.

If you would rather not run this alone, that is exactly what our Selection Sprint and Sample Verification do — written shortlist, verified with measured data, published price.