Markerless Adaptive Visual Servoing Based on YOLOv11-OBB Oriented Object Detection

This paper presents a markerless visual servoing framework that combines YOLOv11-0BB-based oriented object detection with adaptive gain control to achieve precise rotary switch alignment. By directly detecting the switch through its geometric features, the system eliminates the need for fiducial markers and uses the four-corner points as visual features for image-based visual servoing (IBVS), enabling real-time alignment. Furthermore, the adaptive gain strategy dynamically adjusts control parameters based on image error, improving robustness against noise and environmental variations. Experimental validation on a UR5 robotic arm with an Intel RealSense D435 camera has demonstrated superior adaptability and high precision compared to marker-based methods, making the proposed approach well-suited for autonomous substation inspection. Overall, this work enhances automation efficiency and control robustness in dynamic environments, contributing to the advancement of markerless visual servoing technologies.