In this paper, a registration method for outdoor wearable AR systems is described. Our approach is based on using a high precision gyroscope, which can measure 3DOF angle of head direction accurately, but with some drift error. We solved the drift problem with a vision-based drift compensation algorithm, which tracks natural features in the outdoor environment as landmarks from images captured by a camera on an HMD. This paper first describes the detail of the vision-based drift compensation method. Then, a calibration method for the orientation sensor is proposed. Finally, using results from an actual wearable AR system, a comparison of registration error with and without vision-based drift compensation demonstrates the feasibility of the proposed method.