Abstract: A vision/ inertial navigation system (INS) integrated navigation method was proposed to promote the accuracy and autonomy of navigation system for unmanned air vehicle(UAV) autonomous landing. Regarding the INS error equation as process model, and two-view geometry between projective points of identical feature in different instants as measure model, a nonlinear filter was built for INS error estimation. To expedite computation and avoid invalidity of unscented Kalman filter (UKF) algorithm, the square-root-UKF(SR-UKF) was used to estimate the INS error, and the navigation data was compensated by estimating result. Simulation results show that the proposed method is effective to improve navigation system accuracy, and able to reduce the error to 8% of INS.