[Bf-committers] Re: 2d Image to 3d?

GSR - FR famrom at infernal-iceberg.com
Wed Feb 16 19:54:55 CET 2005


Hi,
desoto at blender.spaceisbig.com (2005-02-15 at 2220.17 -0500):
> For once Alex is totally wrong here...
> http://www.graphics.cornell.edu/workshop/talks/acquisition/debevec.pdf
> Alexander Ewering wrote:
> > On Tue, 15 Feb 2005, Austin Benesh wrote:
> >> Would it be too much work to write a sub-routine that takes three 
> >> parts of
> >> an orthographic image (top, bottom, side) and turns them into a mesh? 
> >> Would
> >> it even be possible?
> > What you are asking for is mathematically impossible.

They talk about lots of samples from different angles (not just 3) or
even laser (to get depth directly), so then you can get the reasonable
guess of geometry. They talk about stereo vision to get colour and
depth. They talk about user intervention too to get info about
matching edges (not so auto then). But no three view blueprint (most
probably meaning of the request) to mesh anywhere.

So lets take the car blueprint I suggested (which lacks hidden lines
in the wheels area, btw). A and A' are easy, at least graphically,
dunno how hard would it be in code. Once you have that point in 2
views, you have the XYZ. But given B, could somebody determine B'?
http://www.infernal-iceberg.com/blender/audi-tt-points.png

Probably the best you can get is the cookie cutter solution already
proposed. Maybe with some kind of shading images (lights, helping
pattern) the system could guess but then you are adding extra
conditions. For the general case it is impossible to have something
that can guess, discarding precission errors, the original object
always.

GSR
 
-- 
Food and mails? http://en.wikipedia.org/wiki/TOFU
 


More information about the Bf-committers mailing list