Pose-Estimation Based On Projective Reconstruction For Augmented Reality

Augmented Reality can integrate virtual elements into the user's real surroundings. This characteristic makes AR an exciting technology with a vast potential in human-computer interfaces. Registration of virtual and real objects is a crucial technology for augmented reality.An improved object pose-estimation method based on projective reconstruction is provided in this paper. Using the result of pose estimation, 3D registration can be done for AR systems. The method consists of two steps, including offline reconstruction and online pose estimate.Offline: Matching the keypoints extracted from two reference image. Get the 3D coordinates of the keypoints using projective reconstruction.Online: Tracking the keypoints in each image frames, find out the relative between 3D point and it's project point on the image. Solve the unknown parameter in relative matrix through L-M algorithm.The proposed method has three advantages:(1) Projective reconstruction technology is used before pose-estimation, a marker-less AR system can be made according to this method.(2) It is a precision method because a non-linear least square method is used to minimize the error.(3) The method is robust because the SIFT method it used can find stable feature suit for wide-base line match.The result of the experiment gave out in this paper proved out the method's feasibility and validity.