LPHD: Line Perspective Hexagonal Screw Detection

In industrial production, hexagonal screws are widely used as fasteners, and it is crucial to have fast and accurate inspection of these screws for automated screwing processes. However, this remains a challenge due to various factors such as environmental distractions, lighting, shadows, and screw positioning. To address this, the Line Perspective Hexagonal Screw Detection (LPHD) algorithm has been proposed in this paper, which introduces four innovative techniques: canny value, hexagonal base, perspective correction, and double boundary. The LPHD algorithm starts by feeding the image captured by the robot camera into the YOLO model to obtain the bounding box of the hexagonal screw. Then, the result is subjected to canny edge detection and LSD/FLD line detection. Since the head of a hexagonal screw is a regular hexagonal prism, only three edges are needed to determine a hexagonal contour, which is referred to as the hexagonal base. Perspective correction is proposed to eliminate the camera's perspective effect. If multiple contours are found, a combination of canny value and double boundary is used to identify the optimal result. Ultimately, the LPHD algorithm is applied to find the hexagonal contour in this box. This paper also compares the efficiency and accuracy of LPHD with Mask R-CNN. The results show that LPHD is nearly three times more efficient than the Mask R-CNN algorithm with approximately the same accuracy. Therefore, the LPHD algorithm is a promising solution for detecting hexagonal screws in industrial settings.