Industrial vision is the use of cameras, lighting and image-processing software to inspect, measure, identify or guide products automatically on a production line. A vision system captures an image of each part, analyses it with rules or AI, and sends a pass/fail result or coordinates to a PLC or robot within milliseconds, so every unit is checked.
Key takeaways
- A vision system has five parts: camera, optics, lighting, processing (PC, smart camera or edge device) and software, plus the I/O that talks to the line.
- The four main jobs are inspection (defects), measurement (dimensions), identification (codes and text) and guidance (telling a robot where a part is).
- Lighting and part presentation decide most of the result. Good software cannot fix a poor image.
- Rule-based vision suits measurable features; AI suits variable surfaces and defects that are hard to define.
- The main disadvantages are set-up effort, sensitivity to change, and the need for ongoing calibration and ownership.
What is an industrial vision system?
It is a machine vision set-up built for factory conditions: fixed cameras, controlled lighting and software that makes a decision for every part. Unlike a person, it applies the same criteria on every shift and records the result and image for traceability.
The term is often used interchangeably with machine vision. Computer vision is the wider field of algorithms; industrial or machine vision is its application on production lines, with hard requirements on speed, reliability and integration. See also what machine vision inspection is.
What are the main components?
| Component | What it does | Typical choices |
|---|---|---|
| Camera | Captures the image | Area-scan or line-scan; monochrome or colour; 2D or 3D |
| Optics | Sets field of view, working distance and resolution | Fixed focal, telecentric (for accurate measurement), macro |
| Lighting | Makes the feature or defect visible with high contrast | Ring, bar, backlight, dome (diffuse), coaxial, structured light |
| Processing | Runs the analysis at line speed | Smart camera, industrial PC with GPU, edge device |
| Software | Measures, classifies and decides | Rule-based vision tools, deep learning models, or both |
| I/O and integration | Triggers capture, sends results, rejects parts | Digital I/O, PLC fieldbus, OPC UA, MES/database interfaces |
Two industry standards make components easier to compare and combine. EMVA 1288 defines how camera and sensor specifications are measured and presented, so data sheets from different vendors can be compared. GenICam gives a generic programming interface for cameras, so software can control cameras from different makers in the same way.
How does a vision system work on the line?
Each inspection runs the same cycle in a fraction of a second.
- Trigger: a sensor or encoder signals that a part is in position.
- Acquire: the lighting strobes and the camera captures one or more images.
- Pre-process: software corrects, filters and locates the region of interest.
- Analyse: rule-based tools measure edges, count features or read codes; AI models classify the part or locate defects.
- Decide and act: a pass/fail result, measurement or robot coordinate goes to the PLC, which rejects, sorts or guides.
- Record: results and images are stored for traceability and process analysis.
Processing usually runs at the line for speed. Results and selected images can be sent to the cloud for reporting and model retraining; runtimes such as Azure IoT Edge manage the software on edge devices remotely. Our article on how vision inspection systems work goes deeper.
What are vision systems used for in industry?
- Defect inspection: scratches, cracks, dents, contamination, coating faults and missing features.
- Measurement: lengths, diameters, hole positions, gaps and flushness against tolerances.
- Identification: reading 1D and 2D barcodes, direct part marks and printed text (OCR), and grading code quality.
- Assembly verification: checking that every component, label, cap or seal is present and correct.
- Robot guidance: locating parts for pick-and-place, bin picking and alignment.
- Packaging: label position and content, seal integrity, fill level and print checks in food, beverage and pharmaceutical lines.
2D, 3D or AI: which vision technology fits?
| Technology | Best for | Limits |
|---|---|---|
| Smart sensors | Presence checks, simple single-feature verification | Little flexibility or data |
| 2D rule-based vision | Measurement, code reading, presence and position, consistent defects | Struggles with natural variation and textured surfaces |
| 3D vision (laser triangulation, structured light, time-of-flight) | Height, volume, flatness, bead and sealant checks, bin picking | Higher cost and data volume; reflective surfaces can be hard |
| AI / deep learning | Cosmetic defects on variable surfaces, classification of defect types, natural products | Needs labelled examples, validation and monitoring for drift |
Many lines combine them: 2D tools to locate and measure, and an AI model to judge cosmetic defects. Costs vary widely by technology and integration effort; our guide to vision inspection system costs explains the main drivers.
What are the benefits?
- Every part inspected: automated checks make 100% inspection practical where manual inspection relies on sampling. Statistical lot sampling, described in the NIST/SEMATECH e-Handbook, accepts that some defects pass.
- Consistency: the same criteria apply on every shift.
- Traceability: an image and result per part or batch supports audits and complaint handling.
- Process feedback: defect trends point to tool wear, material changes or machine drift before scrap builds up.
- Safety: people no longer need to inspect in hot, fast or hazardous areas.
What are the disadvantages of industrial machine vision?
The main drawbacks are up-front effort, sensitivity to change and the need for ongoing care. Plan for them rather than discovering them after go-live.
- Set-up and integration effort: optics, lighting, mechanics, PLC integration and validation take engineering time.
- Sensitivity to change: new materials, suppliers, colours or ambient light can shift results. Product changes may need reprogramming or retraining.
- Difficult surfaces: shiny, transparent or highly textured parts need special lighting or 3D techniques.
- Rare defects: AI models need examples; defects that almost never occur are hard to train for.
- False rejects: a system that rejects good parts costs money and gets switched off. Tuning is ongoing.
- Environment: dust, oil, moisture, vibration and temperature affect cameras and lenses. Use enclosures with a suitable IP rating and stable mounting.
- Skills: someone must own the system, its settings and its models after the integrator leaves.
Why is lighting so important?
Because the camera only sees what the light reveals. The goal is maximum contrast between the feature you care about and everything else.
- Backlight: silhouettes for edges, holes and outline measurement.
- Dome or diffuse light: reduces glare on shiny or curved parts.
- Low-angle (dark-field) light: shows scratches, embossing and surface texture.
- Coaxial light: for flat, reflective surfaces such as wafers or polished metal.
- Structured light and lasers: for 3D height and shape.
Shield the station from changing ambient light, and test lighting on real good and bad parts before you choose cameras.
How do you choose an industrial vision system?
- Define the task: defect types, minimum defect size, tolerances, line speed and cycle time per part.
- Collect samples: good parts, real defects and borderline cases, agreed with quality.
- Run a feasibility test: prove lighting, optics and software on the samples before ordering hardware.
- Size the system: resolution from the smallest feature and field of view; processing from line speed.
- Plan integration: triggers, PLC signals, reject mechanism, data storage and MES links.
- Set acceptance criteria: target escape rate and false reject rate for a parallel run.
- Agree support: who maintains, calibrates and retrains, with response times and spare parts.
What maintenance and calibration does a vision system need?
Regular cleaning, scheduled checks with reference parts, and recalibration after any mechanical change. Treat the system like a measuring instrument.
- Daily or per shift: run a golden part (known good) and a known defect part to confirm the system still passes and fails correctly.
- Routine: clean lenses and protective windows; check light output, as LEDs dim with age.
- After changes: recalibrate when cameras, lenses, mounts or lighting are moved or replaced. Measurement systems need calibration with traceable targets.
- Models: review false rejects and escapes, add new examples and retrain AI models on a defined schedule.
- Software: keep versions, settings and models under change control, with backups.
How Opsio helps
Opsio delivers automated visual inspection and computer vision consulting for manufacturers: feasibility tests on your samples, AI model development, edge deployment at the line, and cloud platforms for data, retraining and reporting. For broader questions, see our automated inspection Q&A.
Frequently asked questions
What is the difference between industrial vision and human inspection?
A vision system applies fixed criteria to every part at line speed and records each result. People adapt quickly to new products and judge complex cases well, but get tired and vary between shifts. Most plants combine both: automation for volume, people for exceptions and improvement.
Which industries use machine vision the most?
Automotive, electronics, food and beverage, pharmaceuticals and medical devices, packaging, metals, plastics and logistics. Any process with high volumes, tight tolerances or traceability requirements is a candidate.
Is AI vision better than rule-based machine vision?
Not always. Rule-based tools are simpler to validate and work well for measurement and code reading. AI handles variable surfaces and hard-to-describe defects. The best choice depends on the defect, product variation and how much validation your industry requires.
How small a defect can a vision system detect?
It depends on camera resolution, field of view, optics and contrast. A common rule of thumb is that the smallest feature should cover several pixels. Narrowing the field of view or adding cameras increases the detail you can resolve.
How do you calculate the return on a vision system?
Add up the annual savings: scrap and rework avoided, fewer customer complaints and returns, inspection labour redeployed and fewer line stops. Divide the total project cost by those savings to get the payback period, and test the result with low and high estimates.
Related Guides
Written By
Head of Innovation
Jacob leads innovation at Opsio, specialising in digital transformation, AI, IoT, and cloud-driven solutions that turn complex technology into measurable business value. With nearly 15 years of experience, he works closely with customers to design scalable AI and IoT solutions, streamline delivery processes, and create technology strategies that drive sustainable growth and long-term business impact.
Editorial standards: This article was written by the cloud practitioner named in the byline above. We revise it when the technology or regulation it covers changes; the dates on this page show the latest update. Opsio maintains editorial independence.