Occlusion registration in video-based augmented reality

Augmented Reality overlaps virtual objects on real world. It mixes a real and virtual world which can generate more semantic meanings than either one. To seamlessly merge virtual and real objects, model registration is an important problem. This paper provides a practical approach to do occlusion registration which is a key cue for users to understand the scene. We apply our method to Video-based Augmented Reality where detecting occlusion relationship is challenging because virtual objects are simply superimposed on images of real scenes. By estimating the dense depth of real objects from stereo, results show our approach can efficiently and correctly register virtual and real objects.