Computer vision
Live frames from a camera in the measurement volume, processed for shop lighting, pose variation, and clutter.
Technology
Identity from edges, not from luck.
Conventional image matching likes texture. Machined metal does not provide it. Our stack is built for shape, silhouette, and CAD — so a CMM can know what it is looking at.
Stack
Four layers on a machine you already understand.
CAD-generated views
Solid models produce comparison images. The system is not guessing from a handful of photos of one fixture.
Machine learning
Models emphasize edge and shape cues so textureless parts remain distinguishable — the remaining barrier to self-driven measurement.
The hard problem
Why “just use a camera” has not been enough.
Most vision libraries assume logos, labels, or surface texture. A milled housing is a constellation of edges. We treat that as the signal, then bind identity to the inspection program the CMM should run.
- Robust recognition as lighting and placement change
- Automatic selection and start of the inspection routine
- In-cycle feedback when the measurement looks off
- Path to closed-loop quality between production and inspection
Deployment
Low-cost vision, high-value machine.
The CMM remains the measuring instrument. We add sensing and software so the cell can identify, choose, and run. That is how existing machines become autonomous systems, and how new machines can ship with the option already designed in.
Robotics and handling can complete a lights-out loop; the identification layer is what makes unattended inspection safe to start.
Questions
Short answers
Do we replace the CMM?
No. We automate the decisions around it. Accuracy still comes from your machine and its programs.
What if our parts look similar?
That is exactly the case we design for — families of machined parts that differ in geometry, not paint.
Can this run unattended?
That is the goal of the autonomous workflow: identification, program, measurement, and anomaly checks without a specialist at the keyboard.