The advances of multi-view optical tracking technology led to the rise of image-guided robotic-assisted surgery. The positioning accuracy of multi-view system is extremely crucial in ensuring the safety of the surgery process. The calibration error and camera occlusion may lead to the system switching fluctuation problem, which affects the system stability and limits its usage in situations requiring high safety standards. In this paper, the direct linear transformation (DLT) method is applied to solve multi-view triangulation problem and the influence of the calibration parameter error on the measurement accuracy is analyzed in the semi-physical simulation environment. The results show that the system switching fluctuation has a great influence on the positioning accuracy of the system. In order to improve the measurement stability of the system under camera occlusion, this paper proposes an error compensation strategy based on neural network nonlinear fitting. The measurement data at different locations in the workspace are used as learning samples. The trained network model is used to predict the occlusion points of the system and the measurement results are compensated for errors. Finally, the effectiveness of the compensation method is verified by simulations and real experiments.