Vision-based measurement is widely used in structural monitoring. However, in complex environments such as tunnels, dusty areas, or explosion-proof scenarios, cameras are often installed behind protective windows, which introduce refractive effects that degrade measurement accuracy if conventional air-based imaging models are applied. This study investigates the refractive effects induced by a planar protective layer in an air–protective layer–air medium configuration. A binocular vision measurement model considering the planar protective layer is established based on ray tracing and geometric refraction principles. Simulation experiments are conducted to quantitatively analyse the influence of factors such as target-to-camera distance, protective layer pose, thickness, and camera-to-layer distance on reconstruction accuracy. The results show that neglecting the protective layer leads to significant systematic errors, particularly in the depth direction, while the proposed refraction-aware model effectively eliminates these deviations. Field experiments performed under different environmental conditions further confirm that the corrected reconstruction results are closely consistent with the ground truth and that systematic errors are significantly reduced. The study demonstrates that accurate modelling and calibration of refractive parameters are essential for reliable stereo vision measurements in cross-media environments, providing a practical reference for engineering applications involving protective windows.