Computer Vision System for the Detection of Inclusions in Plastic Components with the Use of 3D Models

The article presents the concept of a system for the detection of inclusions, which are defects appearing as local discolorations on the surface of plastic components manufactured using high-pressure injection molding. This system is a part of a broader optical quality control system for polymer components, which also includes geometric inspection. Inclusions can be effectively detected as areas of non-uniform brightness. However, as the first step the object edges must be excluded from the analyzed image. The edge image is determined based on CAD models of the manufactured components, and the proposed reference pair method ensures precise alignment of the edge mask with the image of the inspected product. The algorithm was implemented in the prototype of a modular quality control system. After appropriate parameter tuning, all inclusions in the test dataset were successfully detected, with the false alarm rate maintained at a level acceptable for industrial applications.