Machine vision is the application of computer vision to the factory: a camera takes images of parts or products, software analyses them, and the system decides and acts, for example by rejecting a defective part. It is used to inspect quality, read codes, measure and guide robots, at rates that human visual inspection cannot sustain.
What is machine vision?
The principle is always the same: detect an object or a feature in a digital image, make a decision from it (good or bad, position, dimension, code read), then send that decision to the line. It is also called industrial vision.
Components of a vision system
- The camera: captures the image. A smart camera also integrates processing in the same housing.
- The optics: the lens, chosen according to part size, distance and required accuracy.
- The lighting: continuous or strobe, it brings out the defect to be detected. It often makes or breaks an application.
- The processing: a board or a computer and image analysis software.
- Triggering and outputs: a sensor signals when to take the image, and the result goes to the controller, the robot or the rejection device.
2D, 3D and other techniques
- 2D vision is the most widespread: appearance inspection, presence or absence, code reading.
- 3D vision also measures relief. Common techniques are laser triangulation, stereo vision, time of flight and structured light.
- Other imaging modes see what the eye cannot: thermography, X-rays, multispectral imaging.
Rules or deep learning
Classic vision works with rules: what distinguishes a good part from a bad one is defined in advance (a dimension, a contrast, a shape). This works very well when the defect is simple and consistent. When defects vary a lot (scratches, stains, the appearance of a natural product), writing every rule becomes impossible.
Deep learning removes this difficulty: the system learns from many images of good and bad parts, then inspects on its own. In return it needs well-chosen, well-labelled images. Poor training data produces false positives, that is good parts being rejected.
Applications on the shop floor
- Quality inspection: surface defects, assembly completeness, label compliance, at high speed and on every part rather than by sampling.
- Code and character reading: barcodes, Data Matrix codes, lot numbers, for traceability.
- Measurement: dimensions and positions, without contact.
- Counting and sorting: by colour, shape or size.
- Robot guidance: the camera tells the robot where the part to pick is.
What manufacturers said at Tech for Industry Show
At Tech for Industry Show 2026, at Paris Expo Porte de Versailles:
- L'Oréal explained that its widely deployed vision systems use AI to detect defects on transparent glass jars, previously very hard to spot, and characterise the defect so it can be corrected faster. Read the article
- Renault cited more than 6,400 vision-based quality checks deployed across its plants. Read the article
- Forvia described on-line problem reporting with automatic defect characterisation by camera, which brings more problems to light and reduces line stoppages. Read the article
- TotalEnergies tested drones fitted with cameras and on-board AI to inspect the ballast tanks of its production vessels and detect corrosion, instead of sending teams inside. Read the article
- NTT DATA showed AI-assisted visual inspection built into the operator's workstation, which frames suspect areas on the image. Read the article
Another example in France
The start-up Visionairy, from the Université Paris-Saclay ecosystem, provides AI visual inspection on production lines; among its first customers it cites the Toshiba plant in Dieppe, in 2020 (Université Paris-Saclay, 03/07/2024).
Market and prices
According to the German mechanical engineering association (VDMA), machine vision revenue fell by 2% in 2025, with components up 4% and systems down 6%; it expects growth of around 3% in 2026.
Smart camera and vision system manufacturers generally quote prices on request: the cost depends on optics, lighting, software and above all integration on the line. Some AI software platforms publish theirs: Roboflow, a platform for developing vision models, offers a free plan and a plan at 39 dollars per month (vendor page, checked on 04/11/2026), excluding industrial integration.
How to succeed with a vision project
1. Define the defect or measurement with the quality team: what is acceptable and what is not, with real examples.
2. Gather parts and images, good and bad, in numbers, under real line conditions.
3. Get lighting and positioning right before choosing the software.
4. Choose between rules and deep learning according to how much defects vary.
5. Test on the line at real speed, and measure false rejects as well as missed defects.
6. Plan the integration with the controller, the robot, the MES and traceability. A specialist vision integrator often handles this part.
Frequently asked questions
What is machine vision, in one sentence?
The use of cameras and image analysis software to automatically inspect, measure, read or guide in a factory.
What is the difference between machine vision and computer vision?
Computer vision is the general field; machine vision is its application to production, with constraints of speed, reliability and integration with the line.
Do you need AI for machine vision?
No. Classic rules are enough for simple, consistent defects. Deep learning helps when defects vary too much to be described by rules.
What role does lighting play?
It brings out the defect. Poor lighting makes an application difficult, whatever the software.
Who installs a vision system?
Often a specialist integrator, who selects the camera, optics, lighting and software, and connects the system to the line.
