Glossary · Machine vision
Pattern matching
Also known as: Template matching, Model-based search
German: Mustervergleich
In image processing, pattern matching is a method that locates a trained reference pattern (template or model) in an image and reports its position, rotation, scale and a match score, used for part location, alignment and robot guidance.
- Machine vision
In one sentence
Pattern matching finds a trained reference pattern in an image and reports position, rotation and match score for location and alignment.
Example
Pattern matching finds each bracket on the conveyor, and its position and angle are sent to the robot for picking.
How it applies
- Engineering: Classic correlation-based matching compares gray values; geometric pattern matching uses edges and is more robust to lighting and partial occlusion. Machine learning methods extend this to variable parts.
- Commissioning: The model is trained on a good part; score thresholds and search ranges are set so that the right part is found reliably and look-alikes are rejected.
- Documentation: The documentation team should document the trained model image, thresholds, search region and the procedure for retraining after a product change, since outdated models cause misplaced picks.
Pattern matching vs. blob analysis
Pattern matching searches for a known shape; Blob analysis segments regions without prior shape knowledge. Pattern matching is preferred when backgrounds are cluttered or parts touch each other.