The observation stage, on which the vision-based monitoring system is installed, experiences slight 6-degree-of-freedom (6-DOF) shaking when working for a long time. As a result, it brings great errors to the measurement results. Therefore, a linear solution method for micro-motion variations of the cameras was proposed. Micro 6-DOF pose variations of the cameras were then linearly calculated with a newly constructed model of two orthogonal cameras in fixed connection. Simulation verification shows that when the three Euler angles of the cameras are less than 30' and the three translation vectors are less than 10 mm, the proposed method delivers better solution accuracy, robustness, and computational efficiency than the typical monocular perspective-n-point (PnP) pose estimation algorithm. Meanwhile, outdoor experiments show that the proposed method better calibrates the measurement error caused by the micro-motion of the observation stage. The subsidence of the target point 44 m away from the observation stage was measured and the result shows that in the case of 6 control points, the average absolute measurement error after the pose variation of the stage is calibrated decreases to 0.2 mm, which verifies the effectiveness and practicability of the proposed method.