面向插植手术机器人的单目视觉空间配准OA

To improve registration accuracy between the workspace of an implantation surgical robot and the surgical environment,a monocular vision-based spatial registration method is proposed.First,the spa-tial pose relationship between the camera coordinate system and the robot coordinate system is established through intrinsic calibration of the monocular camera and hand-eye calibration.Subsequently,a surgical target keypoint detection method based on the CenterNet neural network is developed using deep learning techniques.The Perspective-n-Point(PnP)algorithm is then employed to construct a mapping model be-tween the surgical target coordinate system and the camera coordinate system.To further enhance the sta-bility of spatial registration,the Levenberg-Marquardt(LM)algorithm is applied to perform nonlinear op-timization of the mapping model.By integrating the hand-eye calibration results,an LM-PnP pose map-ping model for the surgical robotic system is established.Finally,experiments are conducted to evaluate the stability and accuracy of the proposed method.The results indicate that the proposed LM-PnP algo-rithm achieves a detection error of less than 0.186 mm for the target circular hole.During robot pose trans-formation tests,the orientation and position deviations of the surgical target range from-1.5° to 1° and-1 mm to 1 mm,respectively.These results demonstrate that the proposed registration method provides high stability and accuracy.Compared with conventional binocular depth camera approaches,the proposed method requires only a monocular camera to achieve precise surgical target registration,operates under nat-ural lighting conditions,and does not rely on infrared sensors or optical marker spheres.