Registration Based on ORB and FREAK Features for Augmented Reality Systems

摘要

This paper proposes a novel registration method for augmented reality (AR) systems based on Oriented FAST and Rotated BRIEF (ORB) and Fast Retina Key-point (FREAK) natural features. In the proposed ORB-FREAK method, feature extraction is implemented based on the combination of ORB and FREAK, and the feature points are matched using Hamming distance. To get good matching points, cross-checks and least median squares are used to perform outlier filtration, and camera pose is esti-mated using the matched points. Finally, AR is rendered. Experiments show that the proposed method improves the speed of registration to be in real time; the proposed method can accurately register the target object under the circumstances of partial occlusion of the object;and it also can overcome the effects of rotation, scale change, ambient light and distance.