In-Situ Vision-Based Detection of Non-Cooperative Markers on Carbon Fiber Reinforced Plastic (CFRP) for Robot End-Effector Positioning

Robot end-effector positioning based on non-cooperative marker detection is a key step in achieving automatic drilling of CFRP, for which a vision-based system offers significant potential. However, challenges arise due to complex background interference, weak texture features and random noise, which complicate the detection of non-cooperative markers. To address these issues, a stepwise robust positioning framework is proposed. First, an enhanced object detection architecture, YOLOv8-PCNet, is developed to improve the detection of small targets and weak texture features, enabling accurate identification of non-cooperative markers. Next, a robust ellipse fitting algorithm, SMCCVC, is developed to achieve high-precision extraction of the center of the circle in noisy environments. Finally, 3D positions are obtained based on binocular vision triangulation, enabling high-precision robot end-effctor positioning. Experimental results show that the proposed method reduces the measurement error of marker positions by more than 55.73% compared to traditional methods, providing an effective technical solution for robotic automated drilling on CFRP components.