Inno Counter — Camera and Image Policy

What the counting cameras record and what happens to it.

What the counting cameras record and what happens to it.

What the camera actually captures

The counting camera watches a production line and captures images of the PRODUCT crossing a line or entering a zone that you defined.

Its purpose is to count units. It is not aimed at people, and the detection model it runs is trained on your product, not on faces.

People working on the line will nonetheless appear in the frame from time to time. That is unavoidable with a camera over a production line, and it is why the workplace notice described below matters.

Face recognition, defect detection, behaviour analysis and attendance are separate features. None of them is part of counting, and none of them runs unless you have separately contracted and enabled it.

Where the images go

Most images never leave your premises. The counting program runs on a PC in your branch, and the images it uses to decide whether something is your product are processed there.

When the model rejects something it was not sure about, the cropped image is kept on that branch PC so it can be reviewed and used to correct the model.

A capped sample of those rejected crops is uploaded so head office can review them centrally: at most one per camera every twenty seconds, only those above a minimum confidence, and at most sixty per camera per day. The cap exists because a single busy line can reject dozens of crossings a minute, and uploading all of them would be neither proportionate nor practical.

A local crop is deleted only once the server confirms it stored it. A refusal keeps it, so nothing is lost by an upload that failed.

Faces

The counting model is trained on your product. It is not a face model and it does not identify people.

Face recognition exists elsewhere in the system, for attendance. Where you enable it, you are processing biometric data, which is special-category personal data under Egyptian Personal Data Protection Law 151 of 2018.

In that case you alone are responsible for obtaining explicit consent from the individuals concerned and for the lawfulness of the processing. We provide the tool.

If a person appears in a counting crop, that crop was kept because of the product in it, not the person. You may ask us to delete any specific image.

Retention and deletion

Rejected crops on the branch PC are cleared as they are reviewed or ignored, and the local store is bounded.

Crops uploaded for central review are kept until they are taught or dismissed, and diagnostic health samples are kept for forty-five (45) days and then removed.

Where a crop is used to teach the model, what survives is the trained model itself. Teaching is how the model improves, and it is described in the Inno Counter terms.

You may ask us in writing to stop keeping images from your line for model improvement.

Notices at the workplace

A camera over a production line is visible processing, and the people working under it must be told.

As the employer you are responsible for putting up a clear notice where the cameras are, saying that the area is monitored by camera, that the purpose is counting production, and who to ask about it.

This is not a formality. It is the difference between lawful monitoring and unlawful monitoring, and the obligation is yours as the controller.

The review queue — what leaves the branch

This repeats the limits stated above in one place, because it is the question customers ask most.

Kept at the branch: the cropped images the model rejected, and the scene they came from, in a bounded local store.

Uploaded: at most one crop per camera every twenty seconds, only above a minimum confidence score, at most sixty per camera per day, and nothing at all once the central queue is full.

Never uploaded: continuous video. There is no continuous recording and no continuous stream retained anywhere.

On a typical line this works out at a few megabytes a day, against the hundreds it would be if everything were sent.

Language and the governing text

These terms were drafted in Arabic. The Arabic text is the authentic and governing version. The English version is provided as a convenience translation, and where the two differ in meaning the Arabic text alone applies.