AI video monitoring for the factory floor, including ESD-protected and cleanroom areas

Most of what is written about "AI visual inspection" covers the first job below. If your problem is about what people do on the floor, it is the second. The table helps you tell which one you are looking for.

If your need is per-unit defect inspection, tell us the part, the defect and the line speed. We will tell you plainly whether area cameras can see it or whether it needs a dedicated station.

Each check below is a rule the plant writes down first: where it applies, what counts as a violation, and who is told. SquintPRO then watches for it continuously instead of relying on spot audits.

• Entry into protected areas. An event when someone enters an ESD-protected or restricted zone, or enters it outside the hours set for that zone.
• ESD and cleanroom requirements. Whether people in the protected area are wearing what the area requires, defined for your area and your gowning rules. Large garments such as smocks and coveralls are easier to see from an area camera than small items such as wrist straps, so test each item on your own cameras.
• Procedure adherence. Where a step in a procedure is visible on camera, it can be defined as a rule, so a deviation shows up as an event with evidence instead of an audit finding weeks later.
• Operational discipline. Events and counts per area and shift, so supervisors can see where the same deviation keeps recurring.
• Safety on the floor. Protective equipment and restricted zones around machinery, on the same platform that handles site safety monitoring.

Interkey Semiconductor, an Interkey group company, assembles network devices in Riyadh, including ONTs, 5G CPEs and mesh Wi-Fi systems. SquintPRO is deployed on that floor, where AI vision supports access and ESD compliance, cleanroom and safety governance, and operational discipline.

That matters for one practical reason: SquintPRO runs on an assembly floor owned by the same company that deploys it for customers.

• Which areas and steps first. Pick one area and two or three checks that matter most, rather than everything at once.
• A written rule for each check. Where it applies, what counts as a violation, and what does not, for example a technician passing through without stopping.
• Camera coverage of the step. A camera has to see the area and the people at a usable distance and angle. A suitability review shows which existing cameras can do this.
• Who receives what. Line supervisor, quality, EHS or the control room, and what each is expected to do with an alert.
• What is kept. Detection records can be kept without retaining continuous footage. Decide retention and access with your quality, EHS and legal owners.
• Where processing runs. On your premises, where the footage does not leave the plant, or in the cloud.
• How the floor is told. Introduce monitoring as a tool for the rules everyone already follows, and say what is and is not recorded. Teams that see it as surveillance tend to work around it.

Start with one area, two or three checks, and success criteria agreed in writing before anything is switched on. Over a period that covers your normal shift patterns, measure:

• Confirmed events as a share of all alerts, to see how much of the alert volume is useful;
• Missed events, found by a person reviewing a sample of footage independently;
• Alert volume per shift, against the number your supervisors agreed they can act on;
• What happened after each confirmed event, to check that alerts lead to action and not just to a log.

Then decide in writing whether to extend, adjust or stop.