IL-STR
Ilmek Akış (Video Analytics)
People counting, density, queue time, PPE (helmet/vest) and security event detection from existing IP cameras.
Vision & Video View
Ilmek Vision
Vision & Video
IL-VIS
Visual quality inspection trained on a few hundred samples: catches surface, assembly and packaging defects without stopping the line.
- Few-shot training
- Anomaly detection
- Edge inference, 15 ms
- PLC reject signal
Traditional machine vision needs months of rules and tuning for every new product. Ilmek Vision uses pretrained visual foundation models and active learning to go live on a new product with 200–500 samples.
Capabilities
- Anomaly detection: train only with good samples and catch unknown defects
- Classification and defect localisation (segmentation)
- Counting, measurement and label/OCR reading
- Under 15 ms inference on edge devices; reject signal to the line PLC
- Image archive and traceability for every product
In automotive supply, packaging, textiles and electronics manufacturing, defect catch rates of 98–99.5% are reached.
How to start
One week of image collection is done at the line, defective and good samples are labelled, and results are observed on the live line in shadow mode (without deciding) for two weeks. If the accuracy target holds, the reject signal is enabled.
| Training data | 200–500 samples |
|---|---|
| Inference | < 15 ms (edge GPU) |
| Line integration | PLC, OPC UA, MQTT |
| Cameras | GigE / USB3 / IP |