Re: correlation of rotated image with perspective

On Jul 31, 8:16 am, serg271 <serg...@xxxxxxxxx> wrote:
On Jul 30, 11:36 pm, Adam Chapman



<adam.chap...@xxxxxxxxxxxxxxxxxxxxxxxx> wrote:
On Jul 30, 9:11 pm, Martin Leese <ple...@xxxxxxxxxxxxxxxxxxxxxxxxxx>
wrote:

Adam Chapman wrote:
Thanks for the reply Bob,

This visual system is designed to go on a military UAV in urban
environments, so anything that lights up will likely be shot at!
I think that a rotated 'H' could be identified usin a foruier-mellin
transform, but it's the distortion from perspective thats really
confusing me.

As Bob said, this is a difficult project to
do with image processing. Given this, any
solution you come up with is unlikely to be
robust.

You need to simplify the problem. Having a
laser on the 'H' sounds good to me. You are
going to need to distinguish your 'H' from
fake 'H's painted by the enemy anyway, and
GPS would seem to be appropriate for this.

If the UAV shoots then, presumably, it will
be fitted with IFF. This can prevent the
thing shooting at the friendly 'H'.

--
Regards,
Martin Leese
E-mail: ple...@xxxxxxxxxxxxxxxxxxxxxxxxxx
Web:http://members.tripod.com/martin_leese/

Thanks for the suggestions.
Ignoring if the 'H' is friendly or not for now, if I could find the
corners of the H shape it would be fairly simple to estimate the
perspective projection, orientation and distance from the landing pad
if the dimesnions of it are known.

I did try finding straight lines with a hough transform, then finding
the equations of those lines and identifying corners as the line
intersections. However, multiple lines were found on the vertical
edges of the H and hence gave me loads of possible corners!

any better ideas?

That is a classical marker registration problem, there are many
solutions for it. Specific solution depend on how noisy image,
resolution and how much calculating power you can spend. First you can
google for "marker registration" "augmented reality" and "mixed
reality". BTW there were some threads about it in this ng.
I myself working with cell phone, so I'm well versed only in low-
calculation-cost low-resolution methods.
For you situation low-cost method would be to use not pure edge-
detector but threshold/adaptive threshold, or combination of threshold-
edge-detector. Idea is extract black connected components and analyze
them. Make contour from the shape. Segment contour into straight lines
(how to do it was discussed in this ng). Now count the number of
straight lines to see if it the same as in H. If test pass run second
test -analyze convexity of the contour, it should be the same as in H
(external lines, internal lines should have the same relationship) .
Now third test - using statistical projective invariants check all the
points of the H. After all test passed you can be reasonably sure you
have identified H. Using corners calculate projective transformation.
Convert to "H" coordinates and make last test - check that all non-
corner points have exactly correct coordinate. Now you have your H and
it's projective transform. The way I described is "cheap" - that is
what I do on cell phone (I dropped statistical projective invariants
though - they were not much help in the low resolution image, but if
your image is high-res they could be extremely helpful).
If you have unlimited calculation power you can use other
approaches - for example there exist projective-invariant variation of
Fourier-Mellin transform (don't ask me about it - never used it and
even hadn't finished the article).

Hi Adam,

maybe you are interested in blob detection, one method that could fit
is Maximally Stable Extremal Regions (MSER), it is fast, robust, it
can handle projective transformation. The "H" will very likely be
detected
as a single mser. Furthermore there has been some work on mser
tracking which can be usefull for you.

J. Matas, O. Chum, M. Urban and T. Pajdla (2002). "Robust wide
baseline stereo from maximally stable extremum regions". British
Machine Vision Conference: pp 384-393.
http://cmp.felk.cvut.cz/~matas/papers/matas-bmvc02.pdf

M. Donoser, H. Bischof (2006). "Efficient Maximally Stable Extremal
Region (MSER) Tracking". CVPR: pp 553-560
http://www.icg.tu-graz.ac.at/pub/pubobjects/docvpr2006

Regards,
Tomas

.

Relevant Pages

  • Re: Obtaining 2d shape from perspective image
    ... information from a 2D perspective projection image? ... "downcast" from homogeneous coordinates to normal coordinates. ... transform is the "divide by w": ... To invert this operation, that is, to get from screen coordinates to ...
    (comp.graphics.algorithms)
  • Re: Normalization XYZ -> xyz
    ... transform. ... My second "suggestion" would be a projection onto a sphere in R^3, again having the same property is mapped to a single point on the sphere), the same holds for the first suggestion. ... is the plane through parallel to X,Z (a plane normal is given ... Again, k*X, k*Y, k*Z delivers the same chromaticities x',y',z' for any ...
    (sci.engr.color)
  • Re: Using a different type of projection matrix in Direct3D?
    ... projection transform with Direct3D apart from the standard canonical ... view transform that you get from the Matrix.PerspectiveFovLHmethod ... You should be able to use any projection ... Would this cause the rendering to flip out, ...
    (microsoft.public.win32.programmer.directx.graphics)
  • Re: correlation of rotated image with perspective
    ... perspective projection, orientation and distance from the landing pad ... I did try finding straight lines with a hough transform, ... resolution and how much calculating power you can spend. ... Make contour from the shape. ...
    (sci.image.processing)
  • Re: correlation of rotated image with perspective
    ... transform, but it's the distortion from perspective thats really ... maybe you are interested in blob detection, ... as a single mser. ...
    (sci.image.processing)