Computer Vision · Edge AI
Computer vision for manufacturing
Cameras, vision servers and gateways that detect events on the shop floor in real time and turn imagery into decisions.
Why computer vision
Camera imagery becomes a production decision
Instead of recordings someone has to review later, you get events labelled with a detection class — ready to raise an alarm, an order or a line stop. Everything is computed at the network edge, without overloading your infrastructure.
Real-time detection
Models analyse every frame the moment it is captured and forward only the events that matter.
Reaction without delay
A detected event instantly triggers an alarm, an order or a line stop through the gateway.
One decision cockpit
Every detection lands in AndonCloud — in one place, next to your line and machine metrics.
System architecture
What the system is built from
Three layers of hardware and software that together turn a video stream into production decisions.
Industrial cameras
A video stream from the shop floor — from a single workstation to entire lines — rugged enough for industrial conditions and matched to a specific detection class.
Vision servers
Vision processors run detection models at the edge, analyse frames in real time and forward only the events that matter, without overloading your infrastructure.
Integration gateways
Gateways connect detections to AndonCloud and to machine controllers, so a detected event instantly triggers an alarm, an order or a line stop.
Data flow
How it works
From a camera frame to a decision in AndonCloud — in four steps, in a fraction of a second.
Image acquisition
Industrial cameras send a continuous stream of frames from workstations and lines.
Inference at the edge
The vision server recognises detection classes locally, without sending recordings to the cloud.
Integration gateway
The event reaches machine controllers and plant systems — instantly.
Decision in AndonCloud
An alarm, an order or a line stop — next to machine metrics on one cockpit.
Detection classes
Ready-made detections for your floor
From safety to logistics — pick the classes that matter on your floor. A few flagship examples below.
Missing hard hat
Entry into a restricted zone
Smoke or fire
Forklift in pedestrian zone
Anomalies
2Occupancy
1Deployment outcomes
Vision that works toward results
−68%
fewer safety events missed by staff
< 200ms
from frame to decision at the network edge
18
ready-made detection classes to choose from
24/7
continuous monitoring without breaks or fatigue
FAQ
Frequently asked questions
- No. Detection models run at the network edge, on a vision server next to the line. Only class-labelled events reach AndonCloud — there is no continuous video stream.
- In many cases yes — we match the configuration to the conditions on your shop floor. If the cameras do not meet a detection class's requirements, we will propose a suitable industrial model.
- Detection is computed locally, so the reaction — an alarm, an order or a line stop — happens immediately after the gateway recognises the event.
- Yes. Beyond the ready-made class catalogue, we train models for the specific scenarios in your plant — from quality control to unusual logistics events.