Knowledge-based machine vision

This paper is concerned with structural matching in a vision system that should obtain the geometrical description of unique workpieces constructed of steel plates. With these data the programming of the robots of an automatic welding system can be completed. The workpieces contain objects from a known set of object types. These objects vary in dimensions, position, and orientation within certain known limits. Several gray level images of the workpieces acquired from different camera positions are being collected. These images are transformed into line segment images derived from the gray-level edges. The object types appear as specific line-structures in line segment images. These line-structures are invariant over a wide range of camera positions because of the assumed class of the objects. A knowledge based system (KBS) is used for the recognition of line-structures in a line image. An inference engine searches for line-structures of object types that are described in a knowledge base. The recognition process is called structural matching.