Zone alerts in industrial settings: use cases, configuration and limits

From recorded footage to useful alerts

On most industrial sites, cameras are used to review footage after something has happened. Zone alerts flip that around: video analytics detects a predefined situation in real time, such as someone in an area where nobody should be, and notifies the right people.

The principle is simple, but how well it works depends almost entirely on configuration. A badly tuned alert fires too often, gets ignored and becomes useless. This article covers the common use cases, the settings that matter and the limits you need to know about.

In most cases, the analysis runs on an edge AI unit installed on site: footage stays on the premises, latency stays low and the system keeps working even if the internet connection drops. Only the event, its timestamp and, where agreed, a context snapshot are sent to the people receiving the alert.

The most common use cases

Zone alerts suit situations that can be described simply in terms of place, duration and time of day:

• restricted or no-go areas: the perimeter around a machine, a storage area for sensitive products, a vehicle manoeuvring area;
• blocked emergency exits or aisles: an object or pallet left for a long time in a space that must stay clear;
• out-of-hours presence: people or vehicles detected in a workshop, construction site or yard outside working hours;
• prolonged stopping: a vehicle standing still in a traffic lane or in front of an access point;
• line crossing: movement in a given direction across a virtual boundary, such as entering a marked-off area.

Setting zones, delays and schedules

Three settings make the difference between a useful system and a noise generator.

The zone itself is drawn on the camera image as a polygon. It should cover exactly the space in question, no more and no less, allowing for perspective. A zone that spills over into a walkway will trigger an alert every time someone legitimately walks past. You can also specify which object classes matter: person, vehicle, object left on the floor.

The delay: being present for one second does not mean the same thing as being present for several minutes. A minimum delay before triggering filters out brief passages and fleeting detections. For a blocked exit, the delay will typically be long; for a hazardous area, shorter.

The schedule: many areas are only off-limits at certain times, such as while a machine is running or outside opening hours. An activation calendar, with planned exceptions for maintenance or cleaning, avoids a constant stream of pointless alerts.

False positives: causes and how to reduce them

No detection system is perfect. False alarms have well-known causes, and it pays to anticipate them:

• changes in light: headlights, reflections on a wet floor, moving shadows, the switch between day and night;
• environmental factors: steam, dust, rain, insects or cobwebs near the lens;
• partial obstructions, which make a person appear and disappear behind a machine or rack;
• a zone too close to the edge of the frame, or a camera too far away for objects to be detected reliably;
• legitimate presence the configuration did not allow for: maintenance crews, contractors, a late delivery.

Cutting the noise without missing what matters

Reducing false positives is a matter of successive adjustments: refining zones, tuning delays, improving lighting or framing, cleaning lenses, adding scheduled exceptions. There is always a trade-off: the more you filter, the fewer false alarms you get, but the greater the risk of missing a real event. That trade-off should be decided with the site managers, not left to default settings.

An observation phase, during which alerts are logged but not sent, lets you gauge the volume and relevance of triggers before rolling them out to teams. You also need to decide who receives each alert, through which channel, and what they are expected to do about it.

Finally, ongoing follow-up is essential: sites change, storage areas move, lighting shifts with the seasons. A periodic review of the alerts triggered, the ones judged unhelpful and the corresponding settings keeps the system credible in the eyes of the teams who rely on it.

A prevention aid, not a certified safety system

This is the most important point. A video analytics zone alert is a tool to support prevention and organisation. It never replaces certified safety devices: light curtains, emergency stops, machine safety sensors, or fire and intrusion detection systems that comply with the applicable standards.

A camera can be covered, knocked out of alignment or go offline; a network can fail; a model can miss a partly hidden person. A video analytics system must therefore never be the only barrier between a person and a hazard. It complements existing measures by giving visibility on recurring situations.

In practice, this means the system’s role should be written down: what it is designed to detect, what it is not, and which certified devices remain the reference. It is also wise to monitor the system itself, with a notification when a camera goes offline or its image becomes unusable.

Informing staff and defining purposes

Zone alerts are about places and situations, not individuals. They must not be used to monitor a particular employee or assess their behaviour, and no facial recognition or biometric identification is needed to detect presence in an area.

Staff must be told about the system and what it is for, and the business, as data controller, must complete the necessary CNDP formalities under Law No. 09-08. To explore possible applications, see our industry and construction pages, along with our Solutions page; we can assess feasibility on your site.