Risk appears in a blind spot
A pedestrian can step into the path of a truck before the operator has any clear line of sight.
Detect the person, read the zone that was drawn there, and trigger the configured warning or action.
RAMS Digital
AI Camera
Industrial vision. Human intelligence.
On-device AI Event-based evidence Works with what you already run
Mounted at the
point of risk.
On the truck, on the cell, above the doorway or over a hazard zone. The mounting position, the field of view and the rules are agreed during the site survey, so the camera watches the place where something actually goes wrong.
On the truckOn the cellAbove the doorwayOver the hazard zone
Intelligence
at the edge.
The picture is read on the device, so detection keeps working when the network doesn’t. What you get back is the event that matters, with its time and zone, not hours of footage to review. It runs off the truck battery or a panel supply, and is set up on site.
Decides on the deviceKeeps working offlineEvents, not footageRules set for your site
Everything it needs.
Nothing it doesn’t.
RAMS Digital
AI Camera
Industrial vision. Human intelligence.
On-device AI Event-based evidence Works with what you already run
The gapRecording is not
Recording is not
the same as knowing.
Traditional cameras are very good at keeping a record. The gap is the distance between something happening and anyone being able to act on it — and that distance is where incidents live. Four situations every floor knows, and what the camera does about each.
Someone enters a controlled area
A robot cell, press area or restricted zone gets entered for maintenance, clearing or a changeover.
Detect the entry, time-stamp it, and pass the event to the local response and the record.
A rule exists only on a signboard
PPE and walkway rules are hard to hold to consistently across every shift, on every aisle.
Monitor the rule continuously and create evidence the moment an exception occurs.
The footage has the answer — afterwards
Hours of recording are rarely much use for the improvement you needed to make on the day.
Keep the event that matters, with its time, its zone and its operational context.
One camera
Every environment.
The same unit bolts to a truck, a fence post, a door frame or a ceiling drop. What changes is the zone you draw and the rule you give it.
On the overhead guard.
Raked toward the floor, facing the direction of travel. It sees the person stepping out from behind the rack before the driver can.
Use casesPut it where
Put it where
the risk is.
Four places it already earns its keep. Each one is the same camera, the same edge AI — just a different line on the floor.
How it worksIt doesn’t just record.
It doesn’t just record.
It responds.
Cameras have watched factories for decades. This one understands what it is looking at, and does something about it before anyone reviews the footage.
- 01 · See
Detect and classify.
People, PPE, vehicles, pallets and machines — picked out of every frame by the NPU inside the housing.
- 02 · Understand
Track, zone, predict.
Where each thing is, where it is heading, and whether it is about to cross a line you drew.
- 03 · Act
Trigger, alert, log.
A stop or slow signal, an alarm, a notification — and a clip of the moment, filed automatically, while the event is still happening.
At the edgeSafety can't wait
Safety can't wait
on your Wi-Fi.
Operational safety cannot depend on a distant server and a perfect signal. Every decision is made on the camera itself; the network is for reporting afterwards, and losing it costs you the analytics, never the detection.
01
Local inference
The image is interpreted on the device, instead of every frame being sent somewhere else for a decision to come back.
02
Event-based evidence
The useful output is the operational event and the context around it — not an endless stream of footage nobody will watch.
03
Site-specific rules
Zones and thresholds are commissioned around your actual floor, because a dock, a robot cell and a forklift aisle do not behave the same way.
04
Connected when available
Events synchronise to RAMS Digital for reporting, trends and cross-system intelligence — when there is a network, and not before.
Build your ownWhat would you
What would you
point it at?
Most use cases are a zone and a rule. If you can describe what should never happen on your floor, the camera can watch for it.
It detects
People, kit and machines.
PersonHelmetHi-vis vestForklift & MHEVehiclePalletZone entryDirection of travelCountDwell timeOperator state
It triggers
Something useful.
Stop or slowLocal alarmBeaconNotificationEvent clipDashboard event
It runs
On-device
inference. Detects and decides without a network, and syncs events when it has one.
Rules, not code
Draw a zone. Say what matters.
IFPerson · no helmetINZone BFOR> 2 sTHENStop signalAlert supervisor
Set at commissioning, tuned from real events — not guessed.
ConveyorsHands near a merge point.
Factory gatesPeople in the vehicle lane.
Overhead cranesAnyone under a suspended load.
Cold storageLow light, fog, reach trucks.
The softwareEvery event,
Every event,
on one screen.
Each camera runs on its own. Connected, they report to RAMS Digital, and individual events become trends — where they repeat, when they happen, and which areas have earned attention. The floor stops being anecdote and becomes something you can take to a safety review.
Events by location · last 30 days
Dock doors 41%Press line 27%Aisles 22%Robot cells 10%
Trend
−62%since install
Latest event
EV-229109:41
TypeNear-miss
SourceMHE-03
ActionSTOPPED
DetailClearance 1.1 m
Interface shown is representative. Figures are illustrative until your own site data is collected.
DeployFrom survey
From survey
to live.
No rip-and-replace. The camera mounts beside what you already have and takes power from what is already there.
- SurveyWe walk the site and find where people and machines actually meet.
- MountA bracket on a guard, a post, a door frame or a ceiling drop.
- Power12–80 V DC through a locking XT30. A few watts.
- CommissionUSB-C on site: draw the zones, set the rules and thresholds.
- LiveEvents start landing. We tune from them in the first weeks.
Tech specs
The sheet.
One unit, one specification — whichever environment it is going into.
Vision
- Optics
- F/1.6 · low-light · anti-glare
- Sensor
- Global shutter · HDR
- Detection
- On-device, at the edge — no cloud in the decision path
- Classes
- Person · PPE · MHE & vehicles · pallet · zone events · operator state
Compute
- Processor
- Edge NPU — sized for every class, every frame
- Inference
- On-device · no platform dependency to detect
- Standalone
- Full detection and response without a network
- Storage
- Rolling event buffer with video evidence
Power & I/O
- Input
- 12–80 V DC · locking XT30
- Service
- USB-C commissioning and diagnostics
- Expansion
- Sealed, field-serviceable bay
- Connectivity
- Wi-Fi to RAMS Digital when in range
Physical
- Housing
- Machined aluminium · integrated fin stack
- Ingress
- IP-rated for wash-down and dust
- Operating temperature
- −20 °C to 50 °C
- Status
- Single status light · fault reporting to dashboard
Figures reflect the published product claims on this page. Certification numbers, exact IP rating, dimensions and weight are to be confirmed against the final production unit.
Where it stopsIt adds a layer.
It adds a layer.
It does not remove one.
Every safety control on your floor today is still there tomorrow. The camera sits on top of them and watches; it takes nothing away, and it is not certified to.
Added
RAMS AI Camera
Detects, reads the zone, responds and keeps the evidence.
sits on top of
Unchanged
- GuardingFences, cages and fixed barriers
- InterlocksGates and doors that stop the machine
- Safety scannersLight curtains and laser scanners
- ProceduresPermits, isolation and safe systems of work
- Operator trainingLicences, refreshers and supervision
- CCTVWhatever is recording today, still recording
Where a camera is wired to a machine response, that integration and its failure behaviour are defined and validated for the specific application before it goes live.
FAQ
Frequently asked questions.
The camera watches the floor, not the person. Continuous footage is never uploaded — it runs a rolling buffer on the device and only keeps the seconds around an actual event. Detection is anonymous: it classifies “person”, it does not identify who. Retention, who can open a clip, and what supervisors see are all set by you at commissioning, and we put that in writing before install.
Start here
Show us where the risk happens.
A truck. A robot cell. A shared doorway. A hazard zone. We walk the operating environment, define what the camera needs to see, and show you how the event becomes a response you can use. You keep the survey either way.