Visual Recognition and Tracking Method for the Scope-Holding Surgical Robot System Based on Instance Segmentation

This paper investigates the feature recognition and servo tracking of metal forceps using instance segmentation results. This paper proposes an instance segmentation algorithm based on YOLO11, which improves the original model by incorporating GSConv, introducing the Convolutional Block Attention Module (CBAM), and replacing the neck module with a lightweight improved BiFPN. These modifications improve segmentation accuracy while largely preserving real-time performance and mitigating the impact of illumination variations on target detection. A servo control law is designed with parameters such as the size of the operating area and servo speed to control the endoscope holder for tracking purposes. An experimental platform for tracking minimally invasive surgical forceps was established, integrating instance segmentation and feature extraction algorithms for tracking experiments. The final results validate the feasibility of this method.