Those questions about how capable the tonchidot solution is - and things like skew and orientation and the like - probably all can be answered if you just look at the SIFT algorithm - as a good baseline example of a computer vision approach:
http://del.icio.us/tag/sift Basically SIFT looks for features that remain consistent in an image despite distortion such as color changes, orientation, skew .... Each "feature" becomes a possible match against its own internal recollection of images... So it doesn't matter if it is 2d or 3d or whatever - it's just fodder. But once it has the match, it can quickly determine the 3d orientation of the image in real space. SIFT is just one approach... (and a particularily bad one - for example it is color-blind among other things). Photosynth takes this to an extreme, and you can also kind of see earthmine and tonchidot and the like as all part of the general trend... You use geolocation to help cull the set... it doesn't do much else. You also try shovel your computation up to a server - but - if you want to stay "pinned" to the image in 3d as you move the camera - you probably need to try to compute your feature keys on the images in real time - which is more than I think the iphone can do right now? - maybe tonchidot has some clever optimizations... our unoptimized code off the shelf showed that it took a second or more ( on a dedicated high performance 1u server at a colo ). See also: http://en.wikipedia.org/wiki/Scale-invariant_feature_transform http://web.engr.oregonstate.edu/~hess/index.html By a photo dns or image dns or image wiki - I mean that there should be a non corporate image metadata repository ... so that when we're all walking around in our digitally augmented realities that we're seeing something that more closely represents "truth" instead of the most egregious spam imaginable... I mean unless you like that stuff. - me
_______________________________________________ Geowanking mailing list [email protected] http://lists.burri.to/mailman/listinfo/geowanking
