Detecting dynamic occlusion in front of static backgrounds for AR scenes

Abstract

Correctly finding and handling occlusion between virtual and real objects in an Augmented Reality scene is essential for achieving visual realism. Here, we present an approach for detecting occlusion of virtual parts of the scene by natural occluders. Our algorithm is based on a graphical model of static backgrounds in the natural surroundings, which has to be acquired beforehand. The design of the approach aims at providing real-time performance and an easy integration into existing AR systems. No assumptions about the shape or color of occluding objects are required. The algorithm has been tested with several graphical models.