Basics
A production-focused explanation of how cameras, lenses, lighting, fixtures and reject logic become a repeatable inspection station.
Direct answer
In production, machine vision is not just a camera and software. It only becomes useful when camera, lens, lighting, trigger, fixture and reject logic repeatedly separate good parts from NG parts under real line conditions.
Sources and editorial review
Last technical review: July 18, 2026 by the Deyi Vision application engineering team.
Industry reference for the image sensor, lighting, lens, processing and communication chain.
Standards context for camera characterization, interfaces and interoperable machine vision systems.
Manufacturing reference for automated visual inspection, measurement, robot guidance and barcode reading.
In production, machine vision only becomes useful when the camera, lens, lighting, trigger, fixture and reject logic can repeatedly separate good parts from NG parts under real line conditions.
Factories often buy a camera first, then discover that glare, part movement, missing trigger timing or loose fixtures prevent stable inspection. The imaging route should be reviewed before the model number.
State whether the system must measure, locate, read, count, accept or reject parts, then send good/NG samples with FOV, working distance and speed.
The camera captures evidence, but the production value comes from repeatability: same part presentation, same light response, same trigger window and the same output rule across normal variation.
A high-resolution sensor cannot rescue glare, shallow depth of field, loose fixturing or a late trigger. Camera, lens, light and mechanical presentation should be reviewed together.
2D vision works from image contrast, edges and printed information. 3D vision adds height, profile or volume data when color or contrast alone cannot separate good and bad parts.
How to test before buying
Test whether the station can make the required output: reject, measure, read, locate or report. Use good, NG and borderline parts under the intended fixture and trigger timing; a sharp image alone is not production proof.
Decision checks
2D uses image contrast; 3D adds height or profile data.
Pass/fail, position, measurement or barcode result.
Prepare 5 inputs: part image, defect, FOV, speed and tolerance.
Decision table
| Factor | Practical rule | RFQ impact |
|---|---|---|
| Primary evidence | 2D uses image contrast; 3D adds height or profile data. | Send sample images for 2D and height tolerance for 3D. |
| Common output | Pass/fail, position, measurement or barcode result. | State the decision the PLC or operator needs. |
| Minimum RFQ inputs | Prepare 5 inputs: part image, defect, FOV, speed and tolerance. | Shortens model selection and avoids catalog-only quoting. |
Application proof
Small connector features require stable contrast, low distortion and repeatable part positioning before a pass/fail threshold can be trusted.
Battery tab weld inspection needs geometry confidence, stable position reference and traceability under high station throughput.
Small parts need fast dimensional verification without slow manual measurement or inconsistent operator positioning.
Common mistakes
Factory handoff
The factory route review starts by checking whether the image can be made stable with lighting and fixture control. Then the camera, lens, reader or 3D sensor route is sized against speed, resolution, interface and installation constraints.
If you already have a Keyence, Cognex, Basler, OPT, LMI, Hikrobot or barcode-reader reference, include it as a reference model. Deyi Vision uses it to understand the application class; final selection still depends on real samples and production limits.
Guide to RFQ
Use the guide to frame the question, then send the details so engineering can recommend a route.
Request engineering RFQGuide FAQ
Machine vision is an automated way to inspect products with cameras, lenses, lighting and software so a production line can measure, locate, accept, reject or trace parts.
The core parts are camera, lens, lighting, trigger or sensor, controller or software, I/O and mechanical fixture. Camera selection is only one part of the system.
Use machine vision when manual inspection is too slow, inconsistent or unable to meet the required measurement, barcode, positioning or defect-detection repeatability.
