MATLAB Central Newsreader - image matching

image matching

Mark Verzeilberg — Tue, 21 Aug 2007 13:29:50 -0400

for our project we want to know if it is possible to find a
image in another image and find the position of our Blimp.

We have a Blimp (kind of a zeppelin) in an old gymanium
flying around taking pictures of the ground to find an
object. We have the idea that we can make some kind of
pattern on the ground so that the blimp can make pictures of
the ground and that they can be matched with a computer made
map with the same pattern as on the ground. This way we want
to determine the position of the blimp. Is this possible
with MatLab and can we get an update rate of at least 1
position per second.

For some more information please ask. The idea is inspired
on DSMAC (Terrain Referenced Navigation).

Re: image matching

Matthew Whitaker — Tue, 21 Aug 2007 21:25:36 -0400

"Mark Verzeilberg" <nlda@mathwork.com> wrote in message
<faepce$o6h$1@fred.mathworks.com>...
> for our project we want to know if it is possible to find
a
> image in another image and find the position of our Blimp.
>
> We have a Blimp (kind of a zeppelin) in an old gymanium
> flying around taking pictures of the ground to find an
> object. We have the idea that we can make some kind of
> pattern on the ground so that the blimp can make pictures
of
> the ground and that they can be matched with a computer
made
> map with the same pattern as on the ground. This way we
want
> to determine the position of the blimp. Is this possible
> with MatLab and can we get an update rate of at least 1
> position per second.
>
> For some more information please ask. The idea is inspired
> on DSMAC (Terrain Referenced Navigation).

Wow,
There are a lot of varaiables here.
1) What resolution would the floor map be at and what would
be the overall dimensions in pixels of the floor map.
2) What height would the blimp be at, what would be the
size of it's field of view and what would the resolution of
the blimp camera be.
3) What is the size of the target object. Is it two
dimensional or 3?
4) Are the blimp images fairly stable or is there a lot of
pitch and yaw effects?

I suppose the first thing to look at would be pattern
correlation where you would take the image from the Blimp
camera and correlate it to the known map. From Gonzalez,
Woods and Eddins - Digital Image Processing using Matlab
they have a small function on page 491 that does this
[M,N] = size(f); %where f would be the map
f = fft2(f);
w = conj(fft2(w,M,N); %where w would be the blimp image
g = real(ifft2(w.*f));
[I,J] = find(g == max(g(:))); %coordinates of best match

Good Luck!

Re: image matching

Dave Robinson — Wed, 22 Aug 2007 08:55:30 -0400

"Matthew Whitaker" <mattlwhitaker@REMOVEgmail.com> wrote in
message <fafl8g$bi7$1@fred.mathworks.com>...
> "Mark Verzeilberg" <nlda@mathwork.com> wrote in message
> <faepce$o6h$1@fred.mathworks.com>...
> > for our project we want to know if it is possible to
find
> a
> > image in another image and find the position of our
Blimp.
> >
> > We have a Blimp (kind of a zeppelin) in an old gymanium
> > flying around taking pictures of the ground to find an
> > object. We have the idea that we can make some kind of
> > pattern on the ground so that the blimp can make
pictures
> of
> > the ground and that they can be matched with a computer
> made
> > map with the same pattern as on the ground. This way we
> want
> > to determine the position of the blimp. Is this possible
> > with MatLab and can we get an update rate of at least 1
> > position per second.
> >
> > For some more information please ask. The idea is
inspired
> > on DSMAC (Terrain Referenced Navigation).
>
> Wow,
> There are a lot of varaiables here.
> 1) What resolution would the floor map be at and what
would
> be the overall dimensions in pixels of the floor map.
> 2) What height would the blimp be at, what would be the
> size of it's field of view and what would the resolution
of
> the blimp camera be.
> 3) What is the size of the target object. Is it two
> dimensional or 3?
> 4) Are the blimp images fairly stable or is there a lot
of
> pitch and yaw effects?
>
>
> I suppose the first thing to look at would be pattern
> correlation where you would take the image from the Blimp
> camera and correlate it to the known map. From Gonzalez,
> Woods and Eddins - Digital Image Processing using Matlab
> they have a small function on page 491 that does this
> [M,N] = size(f); %where f would be the map
> f = fft2(f);
> w = conj(fft2(w,M,N); %where w would be the blimp image
> g = real(ifft2(w.*f));
> [I,J] = find(g == max(g(:))); %coordinates of best match
>
> Good Luck!

Trouble with using fast correlation, is that it will only
work if there is no image scaling between the 'map' and the
blimp image AND they are both have exactly the same
orientation.

I would think that you should read up on image registration
techniques - the IP toolbox has an excellent GUI
demonstration for doing this, however the demonstration
needs manual input to identify matching landmarks on
the 'map' and blimp image. This will provide a
transformation matrix which will allow the blimp image to
be warped to match the 'map', at which point the fast
correlation method can be pushed into action.

Obviously in practice you can't use the manual landmark
selection technique, else you wouldn't have asked your
question. This means that you will need some form of
automation in selecting said landmarks. However without
seeing the sort of image you have for the 'map' and the
blimp image, it is a little difficult to give any advice
here.

Regards

Dave Robinson