Motion Parallax Representation for Indirect Augmented Reality

Indirect augmented reality (IAR) presents pre-generated augmented images for achieving high-quality geometric and photometric registration between pre-captured images and virtual objects. Meanwhile, IAR causes spatial inconsistency between the real world and presented images when users move from the location where the real scene was captured. This paper describes a novel way to address the spatial inconsistency; namely, enabling viewpoint change in IAR. The key idea of this study is to employ an image-based rendering technique using pre-captured multi-view omnidirectional images to provide free-viewpoint navigation. For a pilot study, we have developed an IAR system representing a motion parallax effect using an optical-flow-based camera motion estimation.