Abstract
This research presents a system to construct immersive virtual spaces for human agent interaction which added information for agents not to unnaturally behave by using photos. We need virtual spaces which look like the real world to observe human agent interaction, for expansion of research scope. To make such a space, we think that immersive space, photos and information of objects' positions are needed. We use panorama images to show immersive spaces and depth maps to describe objects' positions. Moreover, we interpolate between panorama images pairs for use to move. To do so, we develop methods which reconstruct depth maps and find corresponding points on panorama images. This research succeeds in creating and implementing a system constructing virtual spaces automatically for HAI in an immersive environment. The system creates automatically a 20m × 20m virtual space in 4 days which looks natural for humans if we shoot around 1200 photos.