Invisible wrote:
>>> Wouldn't it depend on the resolution of the images and the size of
>>> the objects in question?
>>
>> No, it's a simple fact that with only two photos there is never enough
>> information to make a 3D model
> 
> I don't think that's true.
> 

http://www.youtube.com/watch?v=MikG05MV3GU&feature=related

Take any two frames, and tell me if the blue or white side is convex.

> I would have thought it ought to be possible to take a stereo photograph
> and construct a kind of texture-mapped height field from that. I agree
> that if you take a picture of a cup, the computer doesn't really stand a
> chance of figuring out what the *back* of the cup looks like.

Stereo photographs suffer from one problem that the eyes do not appear
to, focus. The brain and eye do all the 'number crunching' to preserve
the information and make you think everything is still in focus,
allowing your eyes to shift and pick up other details. Which also helps
with distance recognition.

To do that with a camera, you would need a ton of pictures or a very
small aperture. The first benefits in adding extra ranger information,
the latter only helps in the amount of data to be processed.

From a picture of a cup, shot edge on in full diffuse lighting, I doubt
a computer could even tell you if it was round, square, or any other
convex or concave shape. Not from just two pictures.

> Now, if you take lots of pictures from lots of different angles... well,
> I'm guessing the hard part is going to be figuring out how the images
> "fix together".

Which is what makes this software look kinda neat. Not only is it doing
that work, but it appears to be doing it in real time.

Post a reply to this message