On 7 Lug, 23:08, Div=F3s P=E9ter <pdi...@[EMAIL PROTECTED]
> wrote:
> On Jul 7, 2:21 am, "cr88192" <cr88...@[EMAIL PROTECTED]
> wrote:
>
>
>
> > "Div=F3s P=E9ter" <pdi...@[EMAIL PROTECTED]
> wrote in message
>
>
>news:9b22721a-9e14-4139-9078-ab4f4e452731@[EMAIL PROTECTED]
>
> > > Hey guys,
>
> > > I am looking for an algorithm to perform inverse 3d projection. What
=
I
> > > mean by inverse 3d projection is the following:
> > > I got perspective 2d projections of single 3d points. In every
> > > projection I know only the 2d coordintes of the points, I can
identif=
y
> > > each point in each projection, but the parameters of the projection
> > > (camera position and angle) are unknown.
> > > It is a bit like microsoft photosynth, but much easier because the
2d
> > > points are identified.
>
> > > If you could please give me some links or keywords about this
topic..=
..
>
> > well, unless you have depth values as well (say, image comes with a
dep=
th
> > map), it isn't really possible to get 3D points from 2D points. if you
=
have
> > this, then it is at least possible to get camera-local 3D points, but
y=
ou
> > would need the camera orientation to get the spatial 3D points back,
..=
..
>
> > > many thanks
> > > Peter
>
> Maybe I was not clear enough, sorry for that. When I said several
> projections I meant projections from different camera positions and
> angles whereby the depth information is encoded in the different
> projection from different depths.
Have a look at Hartley and Zisserman, Multiple View Geometry in
Computer Vision, chapter 19: Auto-calibration.
|
