A flexible novel method of registering virtual objects in monocular AR system is presented in this paper. Monocular AR systems use SLAM-related techniques to obtain the camera pose, of which the translation component is on a random scale. We add a scale calibration process to acquire the scale factor from the SLAM map to the real world and provide a closed-form solution of the transformation between two coordinate systems with different scales. We also describe the framework of an AR system based on our method with implementation. The proposed system can easily initialize virtual objects’ position, orientation and size by using a known reference in the real scene and the reference is no longer needed in the later process. Our method is flexible, simple to set up and easy to control. The results show the proposed method can apply to real-time interactive AR applications.