Knowledge · Machine vision
A label can be present while its print is missing or displaced. In this VisionMaster walkthrough, you load an image, locate the label and examine its dark lettering. Follow the processing steps, then use the test checklist before sending a pass/fail result to the machine.
Build a procedure with three connected steps: Image Source supplies the image, Fast Match locates the label and Blob Analysis examines the dark print. Start with saved images so you can repeat the same input while learning the settings.
The screenshots follow Hikrobot’s label exercise in VisionMaster V4.2.0. Blob Analysis can count dark regions or assess their area. Checking the wording on a label requires text recognition; a region count alone cannot establish that the text is correct.
Create a procedure and add Image Source. For the first exercise, select Local Image and load a still image. This lets you process identical input while exploring the settings. The supplier video uses an image from its local demonstration folder. For your application, use images of the actual parts.
Check that the lettering is in focus and clearly separated from the background. Saturated highlights and motion blur can remove detail before processing begins. Fix the lighting, lens settings and exposure time before tuning thresholds. When you introduce a camera, confirm that live images have the same format and contrast.
Add Fast Match and connect its image input to Image Source. Open the template settings, draw a rectangle around the distinctive label and create the template. Select a feature that remains visible on good parts and contains enough structure to locate reliably. A plain background provides little information.
Run the step and inspect the reported position and match score. Set the permitted displacement and rotation from the way parts are presented. A demonstration threshold is not an acceptance limit for your product. Include an absent label and a similar background that could produce a false match.
Add Blob Analysis. Connect its image input to the source image and use the Fast Match position to place the inspection region. The demonstration connects the fixture input for this purpose. Here, a fixture is a positional reference that lets a region follow the located part.
Draw the region of interest, or ROI, around the relevant print. Use displaced images to verify that it stays on the same lettering. Define the response when Fast Match finds nothing: the inspection must report failed localisation and must not pass on a previous position as a valid result.
For dark lettering on a light background, choose the corresponding polarity, shown as “Dark on Bright” in this version. Adjust the grey-value threshold while watching the detected regions. An overly broad threshold can include shadows and background; an overly narrow threshold can remove thin strokes.
Use area filters to exclude small isolated pixel groups. Then check which letters touch and which split into several regions. This is why blob count cannot simply be treated as character count. Inspect the resulting regions on acceptable and defective samples.
The demonstrated image-processing sequence is the starting point. For your application, add a decision that checks successful localisation and compares the selected measurements with agreed limits. Dark-print area may be useful, provided your test images show that it distinguishes the defect that matters.
Provide a separate error status when an image is missing or processing does not complete. Agree with the machine builder how the result, part identifier and completion signal remain associated. A previous part’s pass result must never be reused for the next part.
Save a versioned solution together with the test images and camera and lighting settings. Close the software, reopen the saved solution and rerun the same images. Confirm that settings and results are retained. Record the software release and required licence as well.
Agree acceptance criteria before tuning. Reserve a separate image set for final evaluation. Record the expected decision, actual decision and explanation of any discrepancy for each image. With a camera connected, measure trigger-to-result time including transfer and communication.
| Test | What should you record? |
|---|---|
| Good parts at different positions | Does the inspection region follow the print and retain the expected pass result? |
| Missing label or missing print | Is the target defect rejected with an identifiable reason? |
| Borderline samples and contrast changes | Which variation is acceptable, and where does rejection begin? |
| Missing image, restart and recipe change | Is an explicit error reported, and is the correct configuration restored? |
Collect representative acceptable and defective parts, the smallest required defect, the available cycle time and permitted product variation. Business-case inputs include current manual inspection time, reject and rework costs, programming, licences and maintenance. The trial must establish which defects are visible and distinguishable; this article claims no detection rate or payback period.
The screenshots come from Hikrobot’s official “Vision Master Basic Operations” video, published on 26 September 2023, showing V4.2.0. This is a supplier demonstration; the additional pass/fail rules and acceptance tests explain how to assess your own application. Menus may differ in other versions. Watch the full video for the demonstrated operation.
Share sample images, camera settings and the time available per part. These provide a practical basis for discussing the trial and computer configuration.
Discuss a feasibility trial