Manufacturing involves many activities that require visual inspection: does the part meet the dimensions, is the label readable, does the quantity match, is the product in the correct position? At high volumes, visual inspection becomes a bottleneck, because people tire and it is difficult to maintain consistent quality with manual inspection. Machine vision makes it possible to automate part of this work. We explain what it is, which tasks it performs in manufacturing and how to assess whether it would benefit your processes.
What is machine vision?
Machine vision is a field of machine learning where computers are trained to interpret the visual world and to identify and classify objects using digital images, videos and deep learning models. In manufacturing, this means that a camera captures an image of a product or process, and the software detects defects, dimensions, text, objects or people from the image. Based on the result, the next action can be triggered without anyone having to look at the image themselves or enter data manually.
In practice, a machine vision system works like this:
- Camera and lighting. An industrial camera is placed at a fixed point in the process, and the lighting is set up so that imaging conditions remain consistent.
- Image. When a product, part or person enters the camera’s field of view, the system captures an image or video frame.
- Software and model. An algorithm analyses the image: it identifies the object, reads text, measures parameters or compares what it sees against predefined requirements.
- The result flows into the management software. The detected data is sent to the management software or ERP, where it becomes a record, a document or a report.
- The next action. Based on the result, the next step in the process is triggered: the part is passed on or set aside, stock levels are updated through the integration and, if necessary, the responsible employee receives a notification.
Simpler applications differentiate shapes, colours and positions to detect defects and categorise objects. More advanced systems can also detect flaws that the human eye cannot distinguish, such as stresses in glass panels or microscopic cracks.
What is machine vision used for in manufacturing?
In manufacturing and warehousing, machine vision is used mainly for the following tasks:
| Task | What the camera does | Where it is used |
|---|---|---|
| Quality control | Inspects defects, dimensions, position and quantity and compares them against requirements | On the production line and in final inspection before dispatch |
| Reading text and codes (OCR) | Reads text, serial numbers and codes from an image and digitalises them | Automating data entry at goods-in and on the production line |
| Object identification and movement tracking | Identifies an object and registers its movement from one position to another | Warehouse management, often together with RFID gates |
| Sorting and assisting robotic production | Differentiates shapes, colours and positions and categorises objects | Automated and robotic manufacturing processes |
| Access management with facial recognition | Identifies a person by their face and grants access | Manufacturing sites and buildings with restricted access zones |
| People counting | Counts visitors to a room | Visitor count monitoring |
Automating quality control reduces the time spent on inspection and human errors. The accuracy of machine vision helps you reach a situation where defective parts do not leave production, and high-quality products are not mistakenly classified as defective. When data from inspection points is brought together in a single system, it can be used to build real-time manufacturing and quality monitoring, i.e. an MES-level solution covering the process from raw material to finished product.
We covered reading text and codes in more depth in a separate article, Computer Vision and OCR in Manufacturing — if your pain point is specifically manual data entry, start there.
Two examples from our projects
Rehvid.com (Vevid AS). Previously, every tyre passport had to be filled in manually on paper, and recurring information had to be re-entered into each product passport. We solved this with OCR-based machine vision: when a tyre is brought in front of the camera, the system reads the data from the tyre, and the passports for tyres coming off the production line are generated automatically. All orders, along with their statuses, are in one environment. The solution covers five workshops and one production unit, and around twenty customers have access to the environment. You can find a more detailed description in the Rehvid.com case study.
Estonian Academy of Security Sciences library. Here, machine vision is used in a different role: it is used to monitor visitor numbers and gain an overview of footfall without manual logging. It is part of a complete solution together with an RFID-based borrowing system and a self-service station. Read more in the Estonian Academy of Security Sciences case study.
These two examples illustrate the range of the same technology: in one, text is read from a product with a complex shape and documents are generated automatically; in the other, people in a room are counted.
When is machine vision a good fit, and what are the limitations?
Machine vision is best suited to places where visual inspection or data reading happens repeatedly and at a fixed point in the process: goods-in, end of line, dispatch, access control. The more consistent the conditions, the more reliable the result.
At the same time, this technology has limitations that need to be stated honestly before implementation.
- Lighting must be consistent. Recognition accuracy depends on whether the object being read is evenly and sufficiently lit. Highly variable light (large windows, moving shadows, differences between day and night shifts) reduces the recognition rate if it is not taken into account during setup. Industrial solutions typically use controlled lighting around the camera.
- The position and orientation of the label in front of the camera matter. The camera reads best when objects arrive in the field of view consistently in the same position. If items arrive in random orientations, either a guiding mechanism or a multi-camera setup is needed. This can be solved, but it adds complexity and cost.
- Worn or damaged labels reduce recognition reliability. The model usually copes with slight wear, but a label that a person cannot read will probably be beyond a machine as well.
It is precisely because of this last limitation that an exception-handling process is an important part of every implementation: it must be clearly agreed what happens when a read fails, and who registers the data in that case, and how.
Machine vision or barcodes and RFID?
These technologies do not exclude each other, because they answer different questions. Machine vision identifies what is visible: a shape, text, a defect, a person. It does not require adding a separate tag to the item. RFID requires a tag, but it does not need a direct line of sight: it is possible to read a tag even when the goods are in a box or packed tightly on a shelf. In practice, they are often combined: for example, in a warehouse, machine vision and RFID gates can be used together to create a complete solution with data that updates in real time. If you are considering tagging items, we compared barcodes, QR codes and RFID in a separate comparison article.
Where to start?
The simplest starting point is to ask: is there a fixed point in your process where repeated visual inspection or manual data entry takes place? If the answer is yes, that is exactly where it is worth starting to assess the need for and benefit of machine vision.
Before choosing the technology, it is worth carrying out a pre-analysis: mapping the current data flow, the physical conditions and the item types, and only then selecting the camera, lighting and software setup. It makes sense to start implementation with a pilot at one point in the process — that way, the value of the solution can be verified on a limited scale before extending it.
If you would like to see how automated visual recognition would fit into your manufacturing, write to us. We will review your process together and give you an assessment of what machine vision can do in your environment and what implementing it requires.
Read more articles
Manufacturing involves many activities that require visual inspection: does the part meet the dimensions, is the label readable, does the quantity match, is the product in the correct position? [...]
In many Estonian manufacturing and trading companies, stock levels are already recorded in some system. Usually it is the stock module of the ERP or accounting software, where purchases [...]
The count is over. The count sheets are on your desk, the book balance is in a file on your computer, and the stocktake report you have to sign [...]
The year-end stocktake has been in the calendar for a long time, yet for a production or warehouse manager it often only becomes real a couple of weeks before [...]
