Engineering
Selecting an Industrial Camera
Camera selection is arithmetic plus a light budget. Do it in this order and the rest of the system gets easier.
Direct answer
How do you calculate the camera resolution needed for an inspection?
Divide the field of view by the smallest feature size, then multiply by the number of pixels you need across that feature. For a 200 mm field of view, a 0.2 mm defect and 3 pixels across the defect, the minimum is 200 / 0.2 × 3 = 3,000 pixels in that axis. Repeat for the other axis and add margin for fixture variation.
Step 1 — Resolution
Required pixels per axis = (FOV in that axis ÷ smallest feature) × pixels per feature. Pixels per feature depends on the task: presence detection tolerates 2-3, character verification wants 4-6, and dimensional measurement with tight tolerance wants 10 or more.
Then add margin. Fixture tolerance, part variation and future SKUs routinely consume 20-30% of a field of view that looked exact on paper.
Fig. — Camera selection dependency chain
01 · Requirement
- Smallest defectmm
- Field of viewmm
- Part speedmm/s
- Cycle rateparts/min
02 · Derived budget
- Pixels across defect3–5 px min
- Sensor resolutionpx × px
- Max exposureµs
- Data rateMB/s
03 · Specified component
- Sensor + camera model
- Interface (USB3 / GigE / CXP)
- Lens focal length + WD
- Illumination geometry
Step 2 — Shutter and exposure against motion
- Global shutter for anything moving; rolling shutter skews moving edges and corrupts measurement.
- Blur (mm) = speed (mm/s) × exposure (s). Keep it well below the smallest feature.
- If exposure must shrink, you need more light or larger pixels, not more resolution.
Step 3 — Interface and bandwidth
Compute the sustained data rate before choosing, and include every camera that shares the host or the switch. Interface choice is the most common late-stage regret in multi-camera stations.
- USB3 Vision
- High throughput, short passive cable runs, single-camera-per-controller bias
- GigE Vision
- Long cable runs, switch-friendly multi-camera, lower per-link bandwidth
- 10GigE
- Higher bandwidth with GigE's cabling advantages, more NIC/CPU load
- CoaXPress / Camera Link
- High-rate line scan and very fast area scan; needs a frame grabber
Step 4 — Lens and sensor format
The lens must cover the sensor format and resolve at the sensor's pixel pitch. A high-resolution sensor behind an under-specified lens produces expensive blur. Confirm mount, working distance, depth of field at the working aperture, and distortion limits if you are measuring.
Frequently asked questions
- Mono or color?
- Monochrome unless color is part of the measurement. Color sensors interpolate, costing effective resolution and sensitivity — often the difference between seeing a low-contrast defect and not.
- How much resolution margin should I add?
- Enough to absorb fixture tolerance, product variation and likely future SKUs — commonly 20-30% on each axis. Excessive margin costs light and bandwidth, so it should be a decision, not a habit.
Related engineering material