System initialization algorithm based on SIFT key points for markerless augmented reality applications

System initialization algorithm applied for markerless augmented reality applications based on SIFT key points is presented. A small number of key-frames of the real environment are calibrated in the offline process and used to detect the target object in video frame. During the on-line initialization stage, the target object is automatically detected by matching the SIFT key points extracted on the input image and the selected key -frames. The RANSAC algorithm is employed to further optimize point correspondences and discard outliers. The camera pose is accurately estimated using an image registration algorithm. Experimental results demonstrate that our algorithm is accurate, robust, and can effectively reduce manual work.