This paper presents a camera tracking system for the spatial stabilization of Augmented Reality (AR) media. Our approach integrates both artificial landmarks (fiducials) and natural features for camera tracking. Artificial landmarks are used for system initialization and computation of initial camera pose. Robust and extendible tracking is achieved by dynamically calibrating the 3D positions of a priori uncalibrated natural features. Analysis and experimental results demonstrate the effectiveness of this approach for presenting stabilized AR media in long camera-motion video sequences.