Back to Case Studies
Manufacturing · Computer Vision

Real-Time Defect Detection System for a Packaging Manufacturer

Computer VisionYOLOEdge AIPythonOpenCV
94%
Defect detection accuracy
60fps
Real-time line processing
8wk
Time to production
The problem

A packaging manufacturer was catching product defects too late — after items were already packed and palletised. Manual spot-checks missed surface defects, misprints, and dimensional errors, and rejects were discovered downstream where they cost far more to pull.

They needed inspection on the line itself — fast enough to keep up with the conveyor, accurate enough to trust, and able to run on the factory floor without a cloud connection.

The constraint that shaped everything

Line-speed inference

Detection had to keep up with a 60fps camera feed on the production line — anything slower would miss defects or bottleneck the line.

Factory-floor reliability

The system had to run on rugged edge hardware on the floor, with no dependence on a stable internet connection.

Low false-reject rate

Over-flagging good product is as costly as missing defects, so the model had to be precise, not just sensitive.

Our approach
1
Defect dataset & labelling
We collected and labelled thousands of line images across defect types — surface scratches, misprints, dimensional and registration errors — under real factory lighting.
2
Custom YOLO model
Trained and tuned a YOLO detector for the specific defect classes, then quantised it to run in real time on edge hardware without sacrificing accuracy.
3
Line-camera integration
Integrated the model with the existing production-line cameras, processing the feed frame-by-frame at 60fps with bounding-box overlays.
4
Reject routing & logging
Confirmed defects trigger an automated reject signal and are logged with an image, timestamp, and defect type for full traceability.
5
Quality dashboard
A dashboard shows defect rates by type, shift, and line — turning inspection data into process insight.
Results in production

The system runs inline at 60fps with 94% defect-detection accuracy, catching defects before product reaches the packing stage. Rework and downstream rejects dropped sharply, and the quality team gained live visibility into defect trends they could act on.

We used to find defects after packing, when they were expensive to deal with. Now they're caught on the line, in real time. The accuracy is better than manual inspection and it never gets tired.
— Operations Director, Packaging Manufacturer
Related service

Computer Vision
on the line.

Real-time, on-device detection for manufacturing and logistics. Catch defects before they ship.