A real-time camera registration algorithm using nature features for augmented reality applications is presented. The system uses a single camera for visual tracking of the nature features extracted from the real scene. A limited number of calibrated key-frames and a rough 3D model of the part of the real environment are required. Accurate camera registration can be achieved by matching the input image and the keyframe, whose viewpoint is as close as possible to each other. The problem of wide baseline correspondence is solved by rendering an intermediate image. Extended Kalman filter is applied for jitter correction. The performance of the algorithm is tested using real image sequences. Experimental results demonstrate that our registration algorithm is accurate and robust, and it can handle significant aspect changes.