Abstract: Objective Aiming at the problem of accurate real-time pose acquisition in autonomous retrieval of USV (unmanned surface vessel), a STag marking visual guidance technology for unmanned vehicle docking and retrieval is proposed. Method Due to the stable attitude characteristics of STag markers, it was selected as a fiducial marker in the visual guidance of this work. By detecting STag markers in the video stream obtained by the camera on the unmanned boat, combined with the camera's internal parameters and the size of STag markers, EPnP and direct linear transformation (DLT) algorithms are fused to calculate the relative pose of the recovery device and the unmanned boat, and amplitude limiting filtering and first-order low-pass filtering are performed to obtain the required lateral offset and heading deviation for LOS docking guidance. Result In the static performance test, the average angular error of target detection is 6.85 °, and the average distance error is 0.056m. In the autonomous recovery guided lake test, the accuracy of static and dynamic docking is within plus or minus 0.5m. Conclusion Compared to traditional unmanned boat docking and retrieval methods, STag marking visual guidance can improve the terminal accuracy of unmanned boat autonomous docking, thereby improving the overall success rate of retrieval.