Worker Entry Detection in Hazardous Zones

How worker entry detection into a hazardous zone works

Detecting a worker entering a hazardous zone means that a computer vision algorithm finds a human figure in the camera image, determines the point where that person contacts the floor and checks whether this point falls inside a previously defined polygon. If it does, the system raises an event: a signal to a beacon or sounder, a controller input, a message to the shift supervisor or an entry in the event log.

The key difference from ordinary motion detection is classification. The system distinguishes a person from a forklift, a pallet, a shadow or a passing vehicle, so the zone does not react to every change in the frame. As a result the zone boundary can run where the hazard actually begins, instead of where a sensor happens to be mountable.

In SafeView.ai this scenario is handled by SPOT - the occupational health and safety event module that runs on streams from existing cameras, deployed on the CORE layer. It requires no rebuild of the CCTV installation and no interference with the machine safety circuits.

person detection and classification in the frame
projection of the ground contact point onto the zone layout
condition check: zone, dwell time, machine state, shift
response: local signal, notification, evidence recording

Zone types: prohibited, warning and conditional

In practice not every entry means the same thing. That is why zones are worth splitting into levels with different response thresholds. The same area can be fully off limits while the machine is running and fully accessible during an inspection.

A warning zone surrounds the prohibited zone and provides an early signal - the person has not yet entered the hazard area but is heading towards it. This layered configuration gives the operator time to react and reduces the number of hard events that have to be reported.

Conditional zones work best where the hazard is intermittent: a robot drop-off area, the space under a suspended load, a crane travel path or the area in front of an open dock door.

prohibited zone - every entry is an event
warning zone - early signal, no escalation
conditional zone - active only in a specific process state
time-based zone - active during set hours or on the night shift

How this differs from light curtains, mats and safety scanners

Light curtains, pressure mats and laser scanners are parts of machine safety systems - they work deterministically, they have defined performance categories and they form part of the stop function. A vision system does not replace these components and should not be treated as a protective device in the sense of a machine risk assessment.

Computer vision complements them where physical guarding is impractical: large open areas, technical walkways, zones with changing geometry and places where people walk around the guards. One camera covers tens of square metres at once and lets you change the zone boundary in the configuration, with no installation work.

The second difference is the evidence layer. A light curtain tells you that the beam was broken. A vision system gives you an event clip, a timestamp and the context needed to establish why someone entered - and whether the problem lies with the person, with the way work is organised or with the zone layout itself.

curtains and mats - safety function, machine stop
computer vision - supervision, early signal, event evidence
one camera covers large and irregular areas
zone boundaries changed without installation work

Setting zone boundaries - what gets decided during deployment

A zone boundary in a vision system is drawn on the image but refers to the real floor. That is why perspective calibration is carried out at commissioning - otherwise a person standing well behind the boundary would be counted as inside the zone, because their head reaches into the polygon.

The reference point is always the contact with the floor, not the whole silhouette outline. On top of that comes a margin resulting from detection accuracy and camera angle - it is worth drawing the zone with a small allowance from the actual hazard edge rather than exactly along it.

Camera placement matters more here than the algorithm itself. A view that is too flat causes occlusions and projection errors, a view that is too vertical shortens the field of observation. In practice the best results come from cameras looking down on the zone at an angle, with a full view of the floor along the boundary.

perspective calibration and projection onto the floor plan
floor contact point as the entry criterion
safety margin along the zone edge
checking for occlusions from racking, machines and loads
verification at night and under changing lighting

How to keep false alarms under control

A system that alarms on every passing walk gets muted within a week. That is why an event should require several conditions to be met at once, not just a line crossing in the image.

The most effective filter is dwell time: briefly clipping the boundary while walking past is not the same as entering and staying inside the zone. The second filter is a link to the process - the zone is active only while the machine is running, the door is open or the crane is moving.

The third mechanism is controlled suppression. An authorised technician, work in inspection mode, scheduled cleaning - such situations should have a defined mode that suspends the zone and records that fact in the log, instead of generating dozens of events that have to be closed manually.

minimum dwell time inside the zone
zone activation tied to machine or process state
shift schedule and maintenance windows
grouping repeated events into a single incident
separate thresholds for warning and prohibited zones

Responding to an event: from signal to event log

Detection without a defined response is just a counter. During deployment the path is agreed: what happens immediately, who is informed and where the event is recorded. The immediate response is usually an audible and visual signal at the point of the event, because that is the only one that reaches the person who has entered.

A notification to the shift supervisor, team leader or health and safety officer makes sense as a second layer, with an event clip rather than a plain text message. The third layer is the event log - a breakdown by zone, shift and hour of day that shows where the problem is systemic.

MOTION adds movement and traffic intensity analysis to this picture, while LOCATE helps tie an event to a specific place on the plant layout. Together they answer not only whether someone entered, but why they keep entering there.

local signal at the point of the event
notification with a clip to the responsible person
controller input or signal to a higher-level system
event log broken down by zone and shift
periodic report for risk assessment review

Step by step deployment and pricing

Deployment starts with a review of the existing cameras and the selection of zones with real risk. A trial configuration is then built on a few points and runs in observation mode - no alarms, only event recording. This makes it possible to verify thresholds before the signalling goes live.

After a few weeks of observation the final boundaries, dwell times and activation conditions are set, and only then are responses switched on. This order protects against a situation where the workforce loses trust in the system in the first few days.

Cost depends on the number of cameras and zones, the condition of the existing installation, the number of integrations and whether the system is only to record events or also to drive signalling. Current options and ranges are available on the pricing page.

camera audit and selection of priority zones
zone configuration and perspective calibration
observation period without alarms
threshold tuning and activation of responses
workforce training and inclusion of events in safety reviews

Frequently asked questions (FAQ)

Can a vision system stop a machine when someone enters the zone?

Technically it is possible to send a signal to a controller or a higher-level system, but a vision system is not a protective device in the sense of a machine risk assessment. It does not replace light curtains, mats or safety scanners. Treat it as a supervision and early warning layer that works alongside the existing safety systems.

Are new cameras needed to detect entry into a zone?

Usually not. SPOT works on streams from the existing installation, provided the camera has a reasonable viewing angle onto the floor along the zone boundary and sufficient resolution in that area. We propose replacing a camera or adding a point only where the view is blocked or the zone sits at the edge of the frame.

How does the system handle people who only walk past the boundary?

That is what the minimum dwell time and the split into warning and prohibited zones are for. Briefly clipping the boundary does not generate a hard event, and someone walking towards the zone may trigger only an early signal. Thresholds are tuned during the observation period, before alarms are switched on.

Is this kind of monitoring GDPR compliant, and can workers be identified?

Zone entry detection does not require facial recognition - the system detects the presence of a person, not their identity. Event data can be limited to a clip and a timestamp, with a retention period agreed with the employer. The scope of monitoring and the purpose of processing must be described in internal documentation and communicated to the workforce.

How many zones can be handled by a single camera?

Several polygons with different levels and activation conditions can be defined in one view, for example a prohibited zone around a machine and a warning zone surrounding it. The limiting factor is the geometry of the view and occlusions rather than the number of zones in the configuration. Details are settled during the camera audit.

Book a free consultation

Detecting worker entry into hazardous zones