Slashdot Mirror
Modeling People and Places With Internet Photo Collections
CowboyRobot writes "Two researchers have created a system that aggregates thousands of photos from around the Web and integrates them into single images. One application is creating maps by taking the GPS coordinates of photos taken from a collection. Another is creating 3D models of historical buildings by automatically pasting together tourists' photos taken from different angles. 'The challenge is that online data sets are largely unstructured and thus require sophisticated algorithms that can organize and extract meaning from noisy data. In our case, this involves developing automated techniques that can find patterns across millions of images.'"
>Another is creating 3D models of historical buildings by automatically pasting together tourists' photos taken from different angles.
I remember seeing a Microsoft (?) thing that did that YEARS ago. Years and years and years ago.
What happened to that?
Highlights:
The first paper I recall was constructing a 3D model of the Berkeley campus from individual snapshots.
Google Street View uses a variant of this technology.
MicroSoft and NASA joined forces after the Columbia accident to generate a view of the Space Shuttle from hundreds of closeup pictures taken from the space station.
I saw a paper by architects constructing the entire interior of an office building from a large series of snapshots. Its considered more accurate for building engineers than the architectural drawings.
It sounds similar to Microsoft's Photosynth, although they may be taking it further.
Photosynth generated point clouds in 3D and let you view each image from its original vantage point, but now seems to be concentrating on panorama stitching. This project seems to have taken a more interesting direction.
Or just take a bunch of photos yourself and upload them to Autodesk's 123d catch and get a 3d model back. It'll stitch together a scene and make a mesh (or a mess, depending on your photo quality) of it.
try { do() || do_not(); } catch (JediException err) { yoda(err); }
Would be cool to recreate the whole real world using this technique. Build an MMO around it. Would be a cool way to crowd source out your game's mapping work.
1) Consolidation of all the still pics taken at an event will be able to be aggregated into a 3D video stream. Any actual video enhances it and adds continuity. And once that becomes an acceptable report of the event, CGI images with artificial timestamps and GPS markers will be constructed to falsify parts of the event, inserting self at location, or removing person from it...whee, the imagination runs riot.
The cylons are coming...
My roommate was working on a system like this at our university 4 years ago. Microsoft was the financing behind the project so I'm sure they've got something a lot more advanced waiting in the wings than what I saw at the time and it was uber badass then.
1. Start up a cloud VM.
2. Feed Streetview images into a 123dcatch instance on it.
3. Run some surface fractal texturing code to apply a "creative" element while cleaning up the surface dirt, pixellation, etc. and reducing file size.
5. Get the mechanical turk to describe the objects for a few cents each.
6. Auction commercial-use licenses to Garmin,
TomTom, etc.
7. Profit $$$
Oh, I'm sorry sir, I thought you were referring to me, Mr. Wensleydale.
Yup and this is the logical next step.
Once you have spatially organised a bunch of photos (what was done in Microsoft's Photosynth tech demo), the next logical step is to use them to create 3D reconstructions.
The researcher (and apparently at least one of them is the same) now build 3D point clouds out of the photos.
Now someone should sell/license the technology to google.
By combining photo albums with the data they already have from street view, they could build some really nice 3D models of towns buildings, monument, and so one.
(AFAIK the current 3D objects they have are still manually done)
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]