An Improved Augmented Reality Registration Method Based on Visual SLAM

Markerless Augmented Reality registration using standard Homography matrix is instable and has low registration accuracy. In this paper, we present a new method to improve the augmented reality registration method based on the Visual Simultaneous Localization and Mapping (VSLAM). We improved the method implemented in ORBSLAM in order to increase stability and accuracy of AR registration. VSLAM algorithm generate 3D scene maps in dynamic camera tracking process. Hence, for AR based on VSLAM utilizes the 3D map of the scene reconstruction to compute the location for virtual object augmentation. In this paper, a Maximum Consistency with Minimum Distance and Robust Z-score (MCMD Z) algorithm is used to perform the planar detection of 3D maps, then the Singular Value Decomposition (SVD) and Lie group are used to calculate the rotation matrix that helps to solve the problem of virtual object orientation. Finally, the method integrates camera poses on the virtual object registration. We show experimental results to demonstrate the robustness and registration accuracy of the method for augmented reality applications.