Abstract: A robust method for determining the relative pose between the camera(vision sensor) and the inertial measurement unit based on the unscented Kalman filter(UKF) is presented.To reduce the impact of gravity acceleration on the calibration result,the method uses iterated Kalman filter to obtain the real-time estimate of gravity acceleration in the inertial sensor coordinate frames.Simulation and experimental results show that,when there are large initial systematic errors or the system is disturbed by serious non-linear noises,the proposed method is still valid and precise for calibrating the 6-DOF relative pose between the visual sensor and the inertial measurement unit.