Home · Services

Computer vision for quality control and industry

Manual visual inspection is slow, doesn't scale, and depends on the concentration of the person doing it. An industrial machine vision system analyses every part, every frame, every image with the same precision, at line speed, without getting tired. The result is fewer defects leaving the plant and fewer costly reworks further down the line.

What a computer vision system for quality detects

The AI visual inspection systems we design are trained on your specific defects, not on generic categories. They work on real images captured directly on the line:

• Surface defects: scratches, cracks, bubbles, stains, inclusions, out-of-tolerance colour variations. Detected in a fraction of a second for every part.
• Dimensional checks: automatic measurement of lengths, diameters, angles and gaps without physical contact, with sub-millimetre precision where the process requires it.
• Assembly control: missing components, wrong orientation, unreadable codes. The system checks every station before the part moves to the next.
• Process anomalies: variations in the production flow, out-of-spec positioning, abnormal conditions detected on the video in real time.

When it's worth investing in machine vision

Not every inspection process makes sense to automate right away. The cases with the fastest ROI are the ones where at least one of these conditions holds:

• High volume and repetitive inspection: thousands of parts a day with the same acceptance criteria. The operator gets tired, the model keeps the same performance.
• Costly downstream defects: a defect that passes inspection and reaches the customer costs orders of magnitude more than one caught on the line. Automated control lowers the escape rate.
• Visual data already available: if you have in-line cameras or historical images of classified defects, model training is much faster and cheaper.
• Traceability required: in sectors where every part needs an inspection log (automotive, medical, food), the system generates the record automatically without manual intervention.

How we work

Every machine vision system is different: lighting, distance, line speed, defect type. We always start from an analysis of the real process before choosing hardware and model architecture:

• Process analysis: we observe the line, collect samples of conforming and defective parts, and measure the constraints on speed, distance and available lighting.
• Dataset and annotation: we build the training dataset with real images of your process. Where defects are rare, we apply data augmentation techniques to balance the samples.
• Training and validation: we train the model, validate it on a separate set and measure precision and recall against your acceptance criteria before going into production.
• Line integration: the system integrates with the PLC, MES or quality management software already in use. No technology island cut off from the rest of the plant.
• Monitoring and drift detection: production conditions change over time. We monitor performance and update the model when the metrics drop below threshold.

Why Latentia

We build machine vision systems that work in the real world, not just in demos with clean samples and controlled lighting:

• EU data and GDPR: production images stay on your own infrastructure or in European data centres. No unnecessary export.
• No lock-in: we deliver the trained model, the integration code and the technical documentation. You can manage the system yourself or hand us ongoing monitoring.
• Made in Italy: in-house team in Naples, no outsourcing. We can support on-site installation in Italy and abroad.
• Response within 24 hours: every enquiry gets a reply the next working day.

Read more