YOLO-based Detection of Drilled Blind Holes in Laminated Panels with Template Similarity Assessment

This paper proposes an end-to-end approach for detecting drilled holes in high-resolution images of furniture panels and verifying their placement accuracy using a computationally efficient algorithm. The method utilizes YOLO object detection and a sliding-window inspection to analyze line-scanned images for any blind holes and corners in panels. The center point coordinates of each detection bounding box are used to create a vector mesh that connects all detected holes and corners. Euclidean distances of these vectors are then calculated, sorted out, and used as a board holes signature. The resulting signature is compared with template measurements to validate their consistency with expected values. The proposed approach achieves high accuracy in detecting drilled holes and corners in furniture panels, with a mean average precision (mAP50) of 99.4%. Additionally, the template similarity assessment algorithm, along with YOLO detection, can spot deviations of ±10px in hole positions compared to the template using our custom dataset. Higher-resolution images can further decrease the minimum deviation detection. The method offers the potential to streamline the furniture drilled hole inspection process and reduce defective production.