Cameras record into an archive nobody watches
The system notices events itself and reports them instantly
07
Computer vision
The cameras are probably already up — but they record into an archive nobody watches. I turn that stream into a system that notices events by itself and reports them the same second.
- YOLO
- OpenCV
- PyTorch
- OCR
- Edge
- RTSP
- Reply in 15 minutes
- Free estimate
- Pay by milestone
- Code and access are yours
sound familiar?
Sound familiar?
Defects are spotted after the batch has already shipped
Quality control right on the line, at conveyor speed
Time tracking relies on good faith
Face clock-in protected against photos
Video cannot be sent to the cloud
Processing on site — only events leave
what is included
What you get
Quality control
Detection of defects and deviations right on the production line, at conveyor speed.
Counting and tracking
How many people came in, where they went, where they cluster. Product counting and shelf compliance.
Text recognition
OCR of documents, receipts, delivery notes and licence plates — written straight into your system.
Safety
Monitoring of restricted zones, protective equipment, abandoned objects and incidents, with an instant alert.
Running on site
The models run on modest hardware right where the cameras are — no video sent outside and no bandwidth costs.
Dashboard and alerts
Events roll up into statistics, and anything critical goes straight to the responsible person on Telegram.
230employees on biometrics
case study
Biometrics instead of passes and timesheets
An employee opens the camera, the system recognises their face, asks them to turn their head to rule out a photo, and the working day begins. No passes, no timesheets.
process
How the work goes
01
Data
I review your camera footage and assess what can genuinely be recognised and what cannot.
02
Model
I collect and label a dataset, train it for your task and measure accuracy on held-out data.
03
Pilot
We launch on one camera or one line and check the results against reality.
04
Scaling up
I roll it out to the remaining locations and connect the dashboard and notifications.
PythonPyTorchYOLOOpenCVONNXTensorRTFastAPIRTSPDockerGrafana
questions
Common questions
Do not see yours? Write to me — I answer within 15 minutes.
Ask a question→Usually yes. What is needed is RTSP access to the stream and enough resolution in the area of interest. I assess it from your footage before any work starts — sometimes moving the camera is enough.
Not necessarily. I can deploy everything on site: processing runs on a local device and only events and statistics leave it. For restricted facilities that is usually a hard requirement.
It depends on the task and the footage quality — I give an honest number after testing on your data. Promising 99% before seeing the material would be a lie.
It depends on the task, the number of cameras and whether a ready model exists or one has to be trained on your data. I start with a pilot on one camera — that shows the real accuracy before the main investment.
Often not: many tasks run on ordinary IP cameras and an inexpensive computer on site. The biometrics in one of my projects runs on a regular laptop webcam.
Yes: OCR of documents, receipts, delivery notes and licence plates, with the data written into your system automatically.
Shall we discuss your task?
Pick what fits — Telegram opens with a message already typed. Just add a couple of words about the task. The estimate is free, and I reply within 15 minutes.
Message me on Telegram- Reply in 15 minutes
- Free estimate
- Pay by milestone
- Code and access are yours
more