24th
Practical World-Tagging
If you haven’t seen Photosynth, you should check it out. Unfortunately, you can’t run the demo in Linux, but you can watch these videos. It’s a neat concept, and I’m surprised at how well it seems to work. They hint at the possibilities of combining social image tagging with 3d reconstruction, and that is truly cool, but in some ways it’s still limited. For instance, it’s not clear that spontaneously grouping local images by tags alone is possible. Maybe if more images had GPS tags this would work better.
I’ve been reading a lot about monocular SLAM techniques, with an eye towards cyborgization. SLAM allows a single camera to simultaneously map its environment and keep track of its position relative to the environment. The research and code are not too hard to digest. There’s some big problems left, like large open environments, scaling and initialization, but it’s getting there. I especially like Fast-SLAM approaches.
Combining a social Photosynth application with SLAM techniques could be interesting. Something like the correlation method of this paper could help automatically group uploaded landmarks from many HMD-equipped stringers. Users with HMDs could tag physical locations and these would become available to other cyborgs. It would appear as if the physical world were sticky with information. Objects would seem to remember their own history, in the form of tags, pictures, and videos of events. Authors and filmmakers could localize fictional events in a real setting, so they could be physically accessed by viewers.
Another benefit of this is it would strengthen the AR-VR link. Pictures could be mapped to feature points with SLAM and produce a fully-textured 3d model available to VR browsers. The online ‘clone’ of the physical space might contain entirely virtual elements, which could also interact in real-space via augmented reality. So you’d have real elements in virtual space, and virtual elements in real space. I always hated that distinction anyway.
