NEVICO

Vision AI · Eindhoven, NL

Vision AI that looks at video and understands it

Our specialty is processes where people and machines work together. The software follows such a process through cameras, turns what it sees into data, and can act on the machine when something goes wrong. It runs on a computer on site.

  • Understands video, over time
  • Specialty processes with people and machines
  • Runs locally, on site
  • Acts on the machine
01

It follows a process, not a single picture

Most camera inspection judges one image at a time: is this product good or bad. Our models follow what happens over time. They know which step of the cycle the machine is in, what normally comes next, and how long each step normally takes.

That matters because many problems only show up in sequence. A slightly turned box is often harmless. The same box at the moment the gripper comes down can cause a crash. The model can tell these apart because it knows where in the cycle the machine is.

People are part of the process too. An operator who steps into the cell to clear a product is normal work, and the model follows that the same way it follows the machine.

02

It turns video into data

A camera produces footage. The software turns that footage into events: a pallet entered, a layer was placed, a pickup started, a person stepped into the cell. Each event has a time, a camera, and the clip it came from.

The result is a record of the process that you can count, search, and compare between shifts. The video stays attached as evidence, so a number in a report can always be traced back to what happened on the line.

03

It runs locally

Everything runs on one computer next to the line. Video is analysed there. The system keeps working when the internet connection drops, and its reaction time does not depend on a cloud service.

Each camera has its own model, trained on images from that line. A model that was trained on one line is not reused to judge another.

04

It can act in the physical world

The computer is connected to the machine's controller (PLC). It reads the machine's signals and it can write to them. When the model sees a bad pickup or a layer that will not hold, it can pause the line or stop the robot before damage occurs.

The connection works in both directions. A machine signal can also tell the software when to capture an image, so the camera looks at the right moment in the cycle.

05

It gets data out of the factory

Machine signals, sensor readings, and video events come together on site, on one timeline. What is needed outside the factory, such as counts, events, and reports, leaves over an encrypted connection. It can go into the tools a team already uses, so nobody has to open a separate dashboard to see it.

Raw footage stays on site. Images that people review to improve the models are sent separately.

06

It keeps learning from the line

Images the model is unsure about go to a review queue. A person labels them, and that camera's model is retrained on the result. Products, packaging, and lighting change, and the model changes with them.

How it fits together

One computer on site, between the cameras and the machine

In

Camerasvideo of the process
MachinePLC signals, sensors

Computer on site

Vision AIfollows the process over time, one model per camera
Events and verdictswhat happened, when, on which camera, with the clip

Out

Actionpause the line, stop the robot
Datacounts, events, reports, over an encrypted connection

Where it is used

Product

Next Vision Copilot

Copilot monitors a production line and stops it when issues occur, preventing damage. Cameras follow the process and Copilot learns how the line normally runs. It is in use on palletizers and robot cells in food and packaging plants.

next-vision.co

Integration in development

The team's daily stand-up

Production teams use software for their daily stand-up and improvement actions. We are building an integration that sends what the vision AI saw on the line to that software as cards.

A card is one event in the team's own words: a bad layer that was stopped, with its photo, or a period in which the line ran in bypass. The team sees it the next morning in the tool and the meeting they already use.

Contact

Ask us about your process

Tell us which process you want to follow and what you want to know about it. We reply by email.