Background Continuous curvilinear capsulorhexis (CCC) requires surgeons to manipulate fragile eye tissue at the microscale. The limited perceptual accuracy of surgeons makes it difficult to precisely position the forceps. Robot technology provides a feasible way to improve the performance of CCC. Methods To provide real-time and high-precision forceps position feedback to the robot, and extract the expert's operation trajectory for robotic autonomous or assisted capsulorhexis, this paper proposes a forceps tracking method based on Kalman filters to precisely position the forceps. By fusing the inertial sensor data and the visual observation data, the proposed method can obtain forceps depth information and decrease cumulative errors in the inertial sensor data. Results Experimental validation on an eye phantom is performed to verify the effectiveness of the proposed forceps tracking method. The experimental results show that the proposed method can track the forceps tip at an average processing speed of 62.8 frames per second, and locate the forceps tip with an acceptable accuracy (within 25 mu m). Then, we use the proposed method to capture the operation trajectory of an expert for robot preoperative trajectory planning. Conclusions The evaluation experiments indicate that the proposed method can accurately locate the tool tip, and efficiently extract the expert's operation trajectory.