Slashdot Mirror
3D Mars Scenes Recreated From Photos
Hoon Mihn Fao links to this BBC article about Microsoft Research scientists generating 3D models based on pictures beamed back from the Mars rovers. It begins "Using techniques originally applied to paintings, the researchers have written software that works out what flat images would look like from different viewpoints. The result is a navigable 3D model created from one or more pictures snapped by the Mars buggies."
(insert favorite FPS game here) map?
To repeat myself, my friend Robert Clemenzi has put up a page on Mars in stereo and has written a program, that can be downloaded, that allows you to study the pictures from the rovers in either cross-eyed stereo or over-under. http://www.cpcug.org/user/clemenzi/science/MarsIma ges_3D.html
Wow. They said something to the effect of being able to see other objects that was not completely visible in the image. "We work out what's behind an object and how we can automatically fill in what's not seen by cameras or painted." I wonder if this would allow you to rotate around objects that are halfway hidden, but what about asymmetrical objects? How can they extrapolate what's not known? Interesting none the less, but where's the video they talked about?
A coworker of mine and I wrote a script to download and generate every possible 3D image from NASA's MER website. It goes through and finds all matching left and right images then makes them into an anaglyph.
If you've got your red/blue glasses you can see them here. We update them with new images every morning. Some of the images are useless but there are also a bunch that NASA never generates for us. They are separated by rover and Sol.
Paul
Lasciate ogne speranza, voi ch'intrate
Would this technique work, for, say, objects hidden or obscured by some sort of flimsy semi-transparent thin fibrous material? Or only hard, rock-like substances? ^^
Grandparent said:
Parent said:
Try understanding what you read. :) Making a stereoscopic view from a single image is a step backwards. They're interpolating to "resolve" information that doesn't really exist. It's like when crime dramas zoom in about 10x and then magically clean up the image to see details that were not captured on a security camera! If you want to extract depth, you need two pictures taken from separate locations. Otherwise you're just making educated guesses.
Please don't tell me you honestly think you can tell that w is in front of Y just by processing the picture from camera A (unless you know the exact dimensions of the WXYZ elements). You need the additional information provided by camera B. Consider a picture where you see a tall person and a short building. You might assume the person is standing in front of the building, but what if it's a miniature building in the foreground? Unless you have a 2nd picture, you can't know for sure.
What is non-trivial is generating a stereogram from a single image. In some circumstances only a single image is available and another will never become available. Are you seriously claiming that it's a step backwards to be able to generate a stereogram from a single image?
The research was first applied (AFAIK) to Renaissance oil paintings where, almost by definition, only a single image from a single viewpoint at a single instant in time is available. They managed to produce very convincing 3D models of the scenes and, from them, stereograms.
There are planetary images in existence which are essentially unique. The techniques of Criminisi and Blake allows 3D models and stereograms to be created from those images too.
Paul
Lasciate ogne speranza, voi ch'intrate
This sort of thing isn't new. Panorama Tools has PTInterpolate, which given two images of the same scene taken from different viewpoints creates any intermediate view. I've never tried it but this makes me want to.