Project 02 · Computer vision
Visual inspection that keeps up with the line.
A camera system that inspects every item on a production line, flags defects in real time and learns from operator feedback.
Capture
Industrial cameras photograph every item under controlled lighting.
- Python
- YOLO detection models
- OpenCV
- ONNX Runtime on edge hardware
- MQTT link to the line controller
- Web dashboard
The challenge
A packaging line moves several items per second. Quality checks are done by eye at the end of the line, so most defects are found late, some are missed entirely, and the cause is hard to trace.
Line Sight puts a camera and a detection model on the line to watch it continuously, flag defects the moment they appear and build the data needed to fix the root cause.
What the system does
- 01
Detects and classifies defects such as scratches, missing caps and misprints on every item, in real time.
- 02
Triggers the reject mechanism and logs each decision together with the image evidence.
- 03
Runs on a small computer next to the line, so it works at line speed and without an internet connection.
- 04
Lets operators confirm or correct detections, and uses that feedback to retrain the model.
How it works
From input to decision in five steps.
- 01
Capture
Industrial cameras photograph every item under controlled lighting.
- 02
Detect
A detection model locates each item and classifies defects in a few milliseconds.
- 03
Decide
Confidence thresholds decide between pass, reject and human review.
- 04
Act
The line controller diverts rejects and the dashboard updates live.
- 05
Learn
Operator corrections feed the next training run.
What this shows
- Training a reliable detector from a few hundred labelled images, with careful augmentation.
- Edge deployment with real latency budgets.
- Closing the loop: monitoring drift and retraining from operator feedback.
Where it applies
- Manufacturing and packaging
- Food processing
- Textiles and garments
- Warehousing and logistics
Contact
Build With Us.
Tell us about the problem you are facing and the data it lives in. We read every message and reply personally.
Prefer email? contact@skvision.mu