Skip to main content
← Ignition & AI insights
Vision + HMI architecture

Computer vision inside Perspective: put visual evidence where operators already work

An AI vision system can detect a defect in milliseconds and still create operational friction if the result lives in a separate application. For sites already using Ignition Perspective, a better pattern is to surface the right visual evidence inside the existing operator workflow while keeping model execution and evidence governance in the vision platform.

22 Aug 20267 min readDelfers EngineeringEngineering perspective
01

The finding needs production context

A bounding box and confidence score are not enough. The useful record connects the image to the product, station, time, work order, recipe or batch that was active when the inspection occurred. That is what allows a quality team to trace a finding forward into disposition and backward into process evidence.

Ignition may already know much of the current station state. MES may know the genealogy. The vision system knows the image, model version and detection result. The integration should join these pieces without making any one system pretend to own all of them.

02

Design the Perspective evidence card around decisions

The operator component should answer a small set of questions quickly: what happened, where, how confident is the system, what image supports the finding, what production object is affected, and what action is allowed next.

A useful component can show the captured image, highlighted region, PASS/REVIEW/FAIL state, confidence, model version, serial or lot, review history and the approved disposition controls. The detailed model-management interface can remain in the computer-vision platform.

  • Image and region of interest
  • Finding and confidence
  • Part / serial / lot / station context
  • Model version and timestamp
  • Review or disposition state
  • Link to full evidence record
03

Design for uncertainty

Production systems must have a defined path for low confidence, unreadable images, camera obstruction and network loss. Those states should be visible to the operator rather than silently converted to pass or fail.

The safest first integration is often advisory: the vision platform returns a finding and evidence, Perspective presents the review, and existing production logic applies the final disposition according to plant rules.

04

Use event-driven integration where it helps

Ignition 8.3 Event Streams provide a project resource for receiving, filtering, transforming and reacting to asynchronous events. Depending on the site architecture, this can be one useful mechanism for moving inspection events into a project without building a polling loop. REST, MQTT, database and other supported integration patterns may also remain appropriate.

The correct interface is the one that fits the customer’s existing architecture, latency requirement and support model.

05

The commercial advantage is simplicity

For an Ignition integrator, this pattern creates a straightforward way to add specialized computer vision without asking the customer to abandon its operator application. For the customer, the new capability appears in a familiar workflow. For the vision platform, the model and evidence lifecycle remain governed in the system designed for them.

That is a better ecosystem story than another standalone dashboard.

The best vision integration does not give the operator another application. It gives the existing application better evidence.
Technical references

Primary sources used for this perspective

  1. Ignition 8.3 User Manual - Perspective Sessions
  2. Ignition 8.3 User Manual - Event Streams
  3. Ignition 8.3 User Manual - Ignition Modules

Product names and trademarks belong to their respective owners. Delfers is an independent technology company; references to Ignition describe interoperability and architecture and do not imply endorsement or certification by Inductive Automation.