In the markerless augmented reality, an improved registration method which combines oriented FAST and rotated BRIEF (ORB) feature and sequential sampling consistency is proposed based on visual simultaneous localization and mapping (VSLAM). The difference between pairs of matching samples is measured through the similarity function, and the samples are extracted from the data subset of the highest quality function. So that the better matching point pairs are obtained and the plane detection is completed in the three-dimensional map reconstruction process. The direction and position of the virtual object is represented by SVD. The registration of the virtual object is completed with the use of pose estimation of cameras. The experimental results show that the proposed method makes the average registration efficiency increases by 34.5%.