An SR-UKF model for integrating INS and Vision navigation system was built to estimate the INS system error, in which regarding the INS system error equation as process model, and homography between two-view as measure model. Aiming at the invalidity of traditional homography between two views when coplanar points exactly in the word coordinate system plane, a novel expression of homography was proposed under the certain circumstances like UAV auto landing based on vision technology. To improve the efficiency and stability of filter, SR-UKF was used for states estimating, and the navigation data was compensated by the estimated result. Simulations show that the method proposed was effective to improve the accuracy of navigation system.