Industrial computer vision on the CCTV you already own
Industrial computer vision is software that watches your existing camera feeds and raises an exception the moment something on site goes wrong. Most plants, mines and depots have far more cameras than people able to watch them, so footage gets reviewed after a loss instead of preventing one. We run detection on the streams you already record, so the gate, the loading bay and the production line are checked every shift and the right supervisor hears about it in time to act.
The footage is recorded. Nobody is watching it live.
A working mine, mill or distribution centre commonly runs dozens of cameras into a control room staffed by one or two operators. Attention on a video wall fades within minutes, so the recording becomes evidence for an investigation rather than a control that stops the incident. Computer vision changes what the cameras are for: the hardware stays where it is, but a machine watches every stream continuously and involves a person only when a rule you set is broken.
What we detect on a plant, yard or gate
Each deployment is built around the loss, delay or safety risk the operation is actually carrying, not around a catalogue of features. The usual requests are protective equipment at plant entrances, people on foot inside exclusion zones, number plates at the gate and weighbridge, verification of what left a dispatch bay, and stoppages on a line.
Built for unstable power and expensive bandwidth
Video is processed on a small edge computer in your own control room, not in a data centre on another continent. No video climbs a metered link, detection carries on when the internet drops, and your footage stays on your premises. The edge machine sits on the same UPS or inverter that already protects your recorder.
A detection nobody acts on is not worth installing
The value sits in what happens in the minute after the system sees something. Each event is routed to where the responsible person already works — a WhatsApp message to the shift supervisor, a job in the maintenance queue, a record against the vehicle in your fleet system — with the image that triggered it attached, so it is confirmed or dismissed on the spot.
Cameras, people and the rules you write down
Watching a workforce is sensitive, and being vague about it is the quickest way to lose the cooperation you need. Before go-live you decide, in writing, which cameras are in scope, what is detected, how long images are kept and who may view them. Where an event feeds a safety or disciplinary process, a person reviews the footage: the system flags, it does not judge.
When a camera is the wrong answer
If a camera cannot physically see the thing, no model fixes that: stock inside a sealed carton, a gauge in the dark, a bearing failing audibly rather than visibly. The honest answer there is a scanner, a sensor or a BLE tag, and you will hear it at the site assessment rather than after signing.
What it covers
Protective equipment and exclusion zones
Flag missing helmets, vests or eye protection at entry points, and people on foot inside marked zones around conveyors, tippers and forklifts.
Gate and weighbridge plate reading
Read number plates at the gate and weighbridge and match them to the expected delivery, so unbooked arrivals and unticketed exits surface the same day.
Dispatch verification
Capture what actually left the bay — vehicle, seal, load, time — so a delivery dispute is settled with a timestamped image, not two recollections.
Line count and stoppage detection
Count units past a fixed point and detect when a line has stopped, so downtime is logged as it happens rather than reconstructed at shift end.
Yard and bay dwell time
Measure how long each vehicle really sits in a bay or queue, turning "the trucks are slow" into a figure per bay, per shift.
How it works
Site assessment (week 0)
We walk the operation with your production and security people, check what each camera can genuinely see, document the recorder, network and power backup, and agree the one or two detections worth starting with.
Integration (weeks 1 to 3)
We connect to the existing feeds over ONVIF and RTSP — the streaming standards nearly every recorder sold here supports — install the edge computer and run in observation mode, where the system logs what it would have flagged but may not yet act.
Rules and guardrails (week 4)
We tune detections against your own footage, including night and dust conditions, then set what counts as an exception, who is notified and what happens when nobody responds. Nuisance alerts are removed here, before anyone's phone starts buzzing.
Go live and optimise (week 5 onward)
The system starts raising live events to the people who own them, and we review weekly at first: what was acted on, what was noise. A first system is typically live in about four to six weeks.
Questions buyers ask
Do we have to buy new cameras?
Usually not. If a camera already produces a picture in which a person could see the thing you want detected, the software can generally work with it. We sometimes recommend re-aiming a camera or adding one at a weighbridge lane, and you hear that at the assessment.
Will it keep working during load shedding or when the fibre is down?
Detection runs on site, so as long as the cameras and edge computer have power it works with no internet at all. Events queue locally and deliver when the link returns. An outage delays the notification, not the record.
How accurate is it in dust, rain and on night shift?
Conditions genuinely affect accuracy, and anyone quoting an accuracy figure before seeing your site is guessing. Night detection depends on your lighting or the camera's infrared range, and heavy dust at a tip point degrades any camera. We test against your own bad-weather footage and say so if a view will not hold up.
Does our video leave the site?
By default it stays on your premises. Only small events leave — a label, a timestamp and, where you approve it, a still image so the supervisor can act. Retention, storage location and viewing rights are agreed before go-live.
Will this replace our guards and supervisors?
No. It replaces the impossible part of their job, which is watching every screen at once. People still handle exceptions and make the decisions; the system makes sure what matters reaches them instead of sitting unwatched on a hard drive.
Find out what your cameras could already be telling you
Book a platform briefing — a 15-minute technical discussion about your site, the cameras and recorder you already have, and the one loss or delay worth attacking first. We are based in Harare and work across Zimbabwe and the wider region. Email sales@eigenstatesystems.com or message us on WhatsApp at +263 77 636 6999.