Microsoft Releases Photosynth
We all heard about Photosynth back in August, but now we can actually start using it. It is mesmerizing to play with and I can totally see this integrated into VE3D (I mean what is one more ActiveX control among friends). Being able to view not only your photographs, but millions of others puts a real spin on Flickr and Zooomr. 2D maps are nice, but this immersion type of experience is the next level.
You can check out all the collections available here and the Photosynth blog.
I guess this is really how those oblique imagery/Pictometry should be navigated…
oh yeah it’s a really cool way of navigating but thats not whats interests me. I’m looking forward to checking out how this actually work with your own collection… there must be some really mean autocorrelation work going on there if it is as non-human-intervention as they claim it to be. Being in the mapping industry for quite some years and seing how auto-correlation really works with well-defined metric cameras, I seriously doubt this is as fully automated as they claim on ordinary digital cameras.
I really hope that the software will be able to export the 3D point cloud that is created. I can imagine that taking a batch of photos of a building, and then being able to export the 3D points for use in many other applications.
Or just think – you could take a video camera and fly over a downtown area – probably a couple of passes to see down to the street between the buildings. Export the images from the video and Photosynth them. You could have a 3D model of all the buildings for use in your favorite 3D earth viewer.
Sure, there are all kinds of coordinate references to setup and transformation of data from the model to the real world – I imagine that would not be too big of a problem (he says, not having tried to program something that complex…)
My work really doesn’t involve creating much 3D data, other that a few objects in AutoCAD for use in CTech’s EVS. But creating a 3D model for me is very time intensive – and I see Photosynth as being able help out with this greatly.
Or am I dreaming?
well another step up would be to create textures for these 3d models from the pictures, I dont really understand why they havent done that yet, as the photos are attached to the 3d space and I guess could be easily “orthorectified” and merged together to represent building facades.
Not sure how you easily go from a pointcloud to a textured model. First the point cloud does not direclty maps to a surface.
Even if you have the surface how do you identify the color of a pixel on that surface. You could project the triangle The same pixel can have different colors in different pictures based on lightning, distance let alone exlcuding a bird thats flying before a wall…
Well as I see it this new application is an extension from a previous class of existing software to create panoramic images. It seems to me this is just an extension of solving the problem of how photographs interrelate into true 3D space. Examples include Hugin, PTGUI, and any other software to create panoramas and Quicktime VR Panos. Thus the problems you describe in your post have already been “solved” (altough not perfectly yet) by these class of software suites.