Resolving Occlusion in Augmented Reality Based on Invariant for Two Views

A new approach to resolving occlusion problem in augmented reality is presented. An invariant for two views is firstly formulated. Based on 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. Further, each contour is labeled as being "behind" or "in front of" in the first two frames, depending on whether it is in front of or behind the virtual object. 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. Two typical experiments in this paper show that this algorithm is feasible for resolving occlusion.