Vision-based natural feature tracking for augmented reality

A new approach is presented to detect and track the natural features of unprepared scenes from sequential images which can be used in the outdoor AR system. Firstly, a corner detection algorithm based on the SUSAN method finds the most reliable feature points in the image of natural scene. Then, those detected points are matched in different image. Furthermore, motion parameters arc calculated by means of epipolar line constraint. Finally, motion prediction is used to minimize the searching scope in the next step. The method is tested through several images of a natural scene from a sequence of video image.