Optical flow tracking enables real-time capture of a spacecraft’s motion trajectory, which is essential for accurate navigation and attitude control. However, existing spacecraft tracking methods primarily focus on whole-object tracking. Optical flow tracking is particularly challenging for non-cooperative spacecraft due to high-speed maneuvers and the complex space environment. To address these challenges, we propose a long-term dense optical flow tracking method based on the Unscented Kalman Filter (UKF), which integrates physical motion equations into optical flow and pose networks to generate and fuse prediction and observation values within the UKF framework. The proposed method is a plug-and-play algorithm that can seamlessly enhance the precision of existing optical flow and pose estimation models in real time. Additionally, recognizing the rigid body nature of spacecraft, we introduce a new structural consistency loss function to constrain the spatial relationships between points, thereby improving the accuracy of optical flow estimation. Furthermore, we design a variance-based outlier correction technique to process deviating observations, minimizing their impact on the overall estimation. Experimental results on SwissCube validate the effectiveness and stability of the proposed method.