Machine vision explained
Machine vision is the use of cameras, lighting, lenses, industrial computers and software to let automated equipment “see” products, parts and processes. In factories and production environments, machine vision systems capture and analyse images to inspect quality, guide robots, read codes, measure dimensions and make fast pass-or-fail decisions. In practical terms, machine vision takes visual judgement and turns it into a repeatable process: instead of relying only on an operator’s eyes, the production line can use a defined inspection rule, a calibrated camera and a reliable process. The result is more consistent inspection, less manual checking and better visibility over what is happening on the line.
Machine vision is closely related to computer vision, but the emphasis is different. Computer vision is the broader field of teaching computers to interpret visual information, and it includes everything from medical imaging to facial recognition and autonomous vehicles. Machine vision focuses on practical, repeatable visual tasks in industrial settings, where speed, reliability, lighting control and integration with machinery matter.
The core parts of a machine vision system
A successful vision project depends on much more than the camera. Each part of the system affects image quality, processing speed and inspection accuracy, so the setup needs to be designed around the task rather than chosen from a generic specification sheet.
Typical machine vision systems include:
Camera: Captures the image. This may be a smart camera with onboard processing, an area scan camera for still images, or a line scan camera for continuous materials and high-speed applications.
Lens: Controls field of view, focus and magnification. The right lens helps the system see the feature that matters without distortion or unwanted background detail.
Lighting: Often the most important element. Good lighting makes edges, textures, colours, codes or defects stand out clearly, while poor lighting can make even advanced software struggle.
Industrial computer or vision controller: Processes images, runs inspection tools and communicates results. More complex applications may need higher processing power, especially where multiple cameras or advanced algorithms are involved.
Software: Performs measurement, pattern matching, character recognition, defect detection or classification. It also defines the pass-fail logic.
Triggering and communication: Sensors, encoders and PLC connections tell the system when to capture an image and what to do with the result.
Mechanical mounting and guarding: Keeps the camera, lens and light stable, protected and correctly positioned in the production environment.
When these parts work together, the system can deliver repeatable inspection even at speeds that would be unrealistic for manual operators.
How does machine vision work in practice?
Machine vision works by controlling how an object is presented, capturing a usable image, analysing that image and sending a result to the production system. The process is usually very fast, but the principle is straightforward: make the important feature visible, compare it with the expected condition and act on the outcome.
A typical sequence looks like this:
The product reaches the inspection point. A sensor, encoder or machine signal tells the vision system that an item is in position.
The light and camera capture an image. Lighting may be timed precisely to freeze motion, reduce glare or highlight a specific feature.
The software locates the part. It may find an edge, shape, mark, code or reference point, even if the item is not perfectly positioned.
Inspection tools analyse the image. The system checks dimensions, presence, orientation, colour, surface defects, printed text or other defined criteria.
A decision is made. The result might be pass, fail, rework, sort into a category, guide a robot or log data for traceability.
The production line responds. A reject gate, robot, PLC, database or operator screen receives the result and acts accordingly.
The best systems are built around the real production conditions. That means considering part variation, vibration, dust, lighting changes, reflections, cycle time and how operators will interact with the equipment.
Common applications of machine vision technology
The applications of machine vision technology are wide-ranging because almost every production process has something that must be checked, counted, measured, identified or positioned. Some tasks are simple, such as confirming a cap is present on a bottle. Others are more complex, such as detecting subtle defects on reflective or textured surfaces.
Common industrial uses include:
Presence and absence checks: Confirming that a component, label, seal, screw, cap or insert is present before the product moves to the next stage.
Defect detection: Finding scratches, dents, contamination, cracks, chips, misprints, stains or surface irregularities.
Dimensional measurement: Checking length, width, diameter, spacing, angle, gap size or alignment against a tolerance.
Code reading and verification: Reading barcodes, QR codes, serial numbers, date codes and batch information for traceability.
Optical character recognition: Reading printed or marked text, such as expiry dates, lot numbers or product identifiers.
Robot guidance: Giving robots coordinates so they can pick, place, assemble, weld, dispense or inspect with greater flexibility.
Counting and sorting: Classifying products by shape, colour, size, orientation or quality status.
Assembly verification: Ensuring that parts have been fitted in the correct position and sequence.
These applications are especially valuable where inspection must be fast, consistent and documented. They can reduce the risk of missed defects, support quality control and free skilled staff from repetitive visual checks.
Machine vision and computer vision are connected, but not identical
The terms are sometimes used interchangeably, but it is useful to separate them when planning a project. Computer vision is the wider technical discipline. It may involve artificial intelligence, image recognition, scene understanding, video analytics and many non-industrial use cases.
Machine vision is usually more tightly defined. It is built for a controlled environment, connected to industrial hardware and judged by whether it can perform reliably on a production line. Success is not just about recognising an image; it is about delivering the right decision at the right time, every time the process runs.
For example, a computer vision model might classify objects in a general photograph. A machine vision system might inspect thousands of identical parts per hour under controlled lighting, reject only the parts that fail a defined rule and store inspection records for quality reporting. Both use visual data, but the operating environment and success criteria are different.
Where machine vision delivers the most value
The value of machine vision becomes especially clear when visual inspection is repetitive, fast, or difficult to perform consistently by eye. On a production line, operators may need to check the same feature hundreds or thousands of times during a shift. Even with experienced staff, fatigue and variation can make consistent inspection difficult. A machine vision system can perform the same check continuously and according to the same defined criteria.
Speed is another important factor. When products move too quickly for reliable manual inspection, cameras can capture images in fractions of a second and software can make a decision before the product reaches the next production stage. This makes machine vision useful not only for finding defects, but also for keeping automated processes moving without adding manual inspection steps.
Machine vision can also make quality control more measurable. A system can record inspection results, measurements and identification data rather than simply passing or rejecting a product. This can help manufacturers spot process variation earlier, improve traceability and understand where problems are occurring.
For automated equipment, visual information can also become an input rather than just an inspection result. Robots may need to know where a component is positioned before picking it up, while assembly equipment may need confirmation that the correct part is in place before continuing. In these situations, machine vision helps the wider production system respond to what is actually happening rather than relying only on fixed positioning or assumptions.
At the same time, machine vision is not magic. If parts are presented unpredictably, lighting cannot be controlled, surfaces are highly variable or the pass-fail rule is unclear, the project needs careful engineering. The earlier these issues are considered, the more likely the final system is to work reliably.
Choosing the right approach for an inspection task
The right machine vision approach depends on what the system needs to see and what decision it needs to make. A simple presence check may only need a smart camera and a dedicated light. A multi-camera inspection cell with defect detection, measurement and traceability may need an industrial computer, specialist optics and more advanced software.
Before selecting hardware, it helps to define the inspection challenge in practical terms:
What feature must be inspected? Be specific: an edge, label, hole, weld, printed code, seal, surface mark or assembled component.
What counts as acceptable or unacceptable? The system needs clear rules, tolerance limits or example images that represent real variation.
How fast is the process? Cycle time affects camera choice, exposure, lighting and processing requirements.
How is the part presented? Position, orientation, movement and vibration all influence the design.
What environment will the equipment face? Dust, moisture, heat, washdown, glare and limited space can all affect component selection.
What happens after inspection? The output may need to trigger a reject, stop a line, guide a robot, update a database or alert an operator.
This discovery stage is where many successful projects are won. By defining the task clearly, teams can avoid overengineering simple checks and underestimating complex ones.
Machine vision in real-world applications
Machine vision is not limited to factory inspection. Industrial computers can also support camera-based monitoring and measurement in demanding environments.
For example, SEBA Hydrometrie uses a MiniDis Fitlet2 as the central unit in a camera-based measurement system. The industrial Mini PC is integrated into the control cabinet and supports data collection and processing.
This example shows how an industrial computer can form part of a larger camera-based system. It also highlights the importance of reliable hardware when equipment must operate in challenging outdoor conditions.
Read the SEBA Hydrometrie case study →
Practical tips for a stronger vision project
A machine vision project is more likely to succeed when image quality, process control and integration are treated as design priorities from the start. Software matters, but it cannot always compensate for a poor view of the part.
Useful steps include:
Start with real samples. Include good parts, bad parts and borderline examples so the inspection criteria are grounded in reality.
Control the lighting. Test different lighting angles, colours and techniques to make the target feature stand out.
Stabilise the part presentation. Consistent positioning reduces complexity and improves repeatability.
Allow for production variation. A system trained or configured only on perfect samples may fail when normal variation appears.
Plan the operator experience. Clear messages, simple controls and useful images make the system easier to support on the shop floor.
Think about maintenance. Lenses may need cleaning, mounts may need checking and inspection limits may need review as products change.
Validate before full rollout. Testing under realistic conditions helps uncover issues before the system becomes business-critical.
These steps keep the focus on dependable performance rather than impressive technology for its own sake.
The future of machine vision is more flexible and data-led
Machine vision continues to evolve as cameras, processors and software become more capable. Traditional rule-based inspection remains highly useful, especially where the task can be clearly defined. At the same time, AI-assisted tools are making it easier to classify complex defects, manage natural variation and handle inspection tasks that were previously difficult to describe with simple rules.
The practical direction is not “AI versus traditional vision”. In many industrial settings, the best approach is a combination: controlled imaging, robust engineering, clear pass-fail logic and advanced algorithms where they genuinely improve performance. As systems become more connected, inspection data can also support process improvement, predictive maintenance and faster root-cause analysis.
For manufacturers, the key is to match the technology to the problem. A reliable simple system is often better than an unnecessarily complex one. The goal is not just to capture images, but to create trustworthy decisions that improve quality and productivity.
Key takeaway
Machine vision systems give industrial equipment the ability to inspect, measure, identify and guide using visual information. They combine cameras, lighting, optics, computers and software to automate decisions that once depended on manual visual checks.
The best results come from clear inspection goals, controlled imaging and practical integration with the wider production process. Whether the application is code reading, defect detection, robot guidance or assembly verification, machine vision works best when it is designed around the real-world task, not just the technology.