Invariant-Based Augmented Reality on Mobile Phones

Abstract

A calibration-free augmented reality based on affine invariant is firstly formulated by tensor method. This approach does not use the calibration parameters of the camera and the 3D locations of the environment’s object, and can realize the augmentation of virtual objects. Meanwhile, a new approach to resolving occlusion problem in augmented reality is presented. Based on an arm-optimized implementation of  the Scale Invariant Feature Transform (SIFT) algorithm developed by David Lowe and Random Sample Consensus (RANSAC) algorithm, the point correspondences in any two views are determined. According to the invariant for two views and these point correspondences, the occluding contours can be transferred to any views, so the occlusion is resolved. Some typical experiments show that the approach of invariant-based augmented reality is feasible on mobile phones.