Markerless Registration Based on Natural Features Tracking for Augmented Reality

In augmented reality system, registration is always a problem which is difficult to resolve effectively, especially for marker less registration. Due to the placed marks for marked visual registration method, it would cause visual pollution, and the registration often fails when the mark is partially occluded. Registered quality directly affects the fusion effects of the virtuality and reality. This paper proposed a marker less tracking registration method combined natural feature points with affine reprojection. Firstly it extracted natural feature points from real scene with SIFT algorithm, then exploited KLT algorithm to track and match the effective feature points in each image, carried out the homography matrix of each frame with affine reprojection technology, and computed the real scene 3D registration for each frame image, so as to implement the tracking registration effectively. This method is also applicable to those cases of plane feature points being occluded.