RSS News Feed

object recogniction from edge map

Message-ID:
Subject:

object recogniction from edge map

Date:Mon, 5 Jan 2009 13:57:08 +0100


Hi,
I am doing machine vision project in Spartan3a dsp 1800a board. I have
completed sobel edge detection in hardware . Now I have boundary of
the single object. My problem is to identify whether the given
boundary is of cube or sphere. I am planning to complete this portion
using microblaze processor . Can any body suggest me simplest
algorithm or methods for this.
Thank you

Message-ID:
Subject:

Re: object recogniction from edge map

Date:Mon, 5 Jan 2009 22:25:15 +0100


On Jan 5, 2:57=A0pm, SUMAN  wrote:
> Hi,
> I am doing machine vision project in Spartan3a dsp 1800a board. I have
> completed sobel edge detection in hardware . Now I have boundary of
> the single object. My problem is to identify whether the given
> boundary is of cube or sphere. I am planning to complete this portion
> using microblaze processor . Can any body suggest me simplest
> algorithm or methods for this.
> Thank you

For sphere you should just detect circle, for cube - hexagon with
parallel sides. Specific depend on you processor bottlenecks - is it
memory, performance, etc.

Message-ID:
Subject:

Re: object recogniction from edge map

Date:Mon, 5 Jan 2009 23:45:47 +0100

serg271 wrote:
> On Jan 5, 2:57 pm, SUMAN  wrote:
...
>> My problem is to identify whether the given
>> boundary is of cube or sphere.

> For sphere you should just detect circle, for cube - hexagon with
> parallel sides. Specific depend on you processor bottlenecks - is it
> memory, performance, etc.

I would say that the most striking difference
between a circle and a hexagon is the pointy
bits.  So you could detect and count the
points.

Zero and its a sphere.  Four or more and its
a cube.  (Four because the general hexagon
can degenerate into a square).  You would
need to examine sample images to decide how
to assign 1, 2, and 3 points.

This still leaves you the problem of defining
a point.

-- 
Regards,
Martin Leese
E-mail: please@see.Web.for.e-mail.INVALID
Web: http://members.tripod.com/martin_leese/

Message-ID:<9b729c1a-4b41-42c6-a372-5bb11b3fde67@q30g2000prq.googlegroups.com>
Subject:

Re: object recogniction from edge map

Date:Tue, 6 Jan 2009 13:26:55 +0100


First of all you can find simple example of solving such problem in
OpenCV. Then you can take a look on Hough transform in general, if
you're going to work with contours as arrays of 2d points. I've seen
an article on dspdesignline.com  about road sign identification
system. They sought circle of road sign with Hough transform.
If you are going to build more complicated system that would be able
to seek much more figures then you should look to contours analyzing.

Message-ID:<0565b5e0-5319-44f7-ac92-9dea5ec48d1d@x16g2000prn.googlegroups.com>
Subject:

Re: object recogniction from edge map

Date:Wed, 7 Jan 2009 10:25:36 +0100

On Jan 5, 12:57=A0pm, SUMAN  wrote:
> Hi,
> Can any body suggest me simplest algorithm or methods for this.
> Thank you

I would try to use least square fitting first and define some sort of
threshold - fit circle/sphere everywhere, but if LS error is larger
than, say radius dependent threshold, then it's a rectangle/cube.
Should be easy to implement I think.
Good luck.

Message-ID:
Subject:

Re: object recogniction from edge map

Date:Wed, 7 Jan 2009 10:34:14 +0100


On Jan 7, 9:25=A0am, "P. Galibarov"  wrote:
> On Jan 5, 12:57=A0pm, SUMAN  wrote:
>
> > Hi,
> > Can any body suggest me simplest algorithm or methods for this.
> > Thank you
>
> I would try to use least square fitting first and define some sort of
> threshold - fit circle/sphere everywhere, but if LS error is larger
> than, say radius dependent threshold, then it's a rectangle/cube.
> Should be easy to implement I think.
> Good luck

Just to clear up my idea a little bit:
fit circle using LS -> find center -> calculate distance from the
center to every point - if exceeds R + alpha*R -> cube, otherwise
sphere

Message-ID:<7d1d50e1-664e-4e0b-adb9-94c2fdccdf49@d42g2000prb.googlegroups.com>
Subject:

Re: object recogniction from edge map

Date:Wed, 7 Jan 2009 16:07:44 +0100


> I would try to use least square fitting first and define some sort of
> threshold - fit circle/sphere everywhere, but if LS error is larger
> than, say radius dependent threshold, then it's a rectangle/cube.
> Should be easy to implement I think.
> Good luck

The way you suggest to try LS algo is quit similar to Hough transform.
But the second one has very wide use practices for such tasks.

Message-ID:<5ce985a7-e688-427c-9bb1-386be2ccfdf5@a12g2000pro.googlegroups.com>
Subject:

Re: object recogniction from edge map

Date:Thu, 8 Jan 2009 09:49:23 +0100



 Once upon a time I read something about "ciruclarity".

 The idea is to use the area and perimeter of the geometrical figure.

 You calculate the area of the figure, and from it, you calculate the
 radius of a circle with that area, Ra.

 Then you do the same with the perimeter: calculate it and then
 calculate the correspondent radius of a circle whit that perimeter,
Rp.

 Finally you calculate Ra/Rp (so you get a no dimensional number,
which is always a good thing
 when comparing geometrical features, and you don't want to calculate
 Rp/Ra in order to avoid posible divisions by zero), and the closer
this value
 gets to 1, the "more circular" is the figure.

 I guess this value is 1 if and only if the figure is a circle, for
the rest of the figures it's always
 bigger or smaller (I don't remember which is the actual case).

Message-ID:
Subject:

Re: object recogniction from edge map

Date:Thu, 8 Jan 2009 21:17:11 +0100


Groupuco wrote:
>  Once upon a time I read something about "ciruclarity".
> 
>  The idea is to use the area and perimeter of the geometrical figure.
> 
>  You calculate the area of the figure, and from it, you calculate the
>  radius of a circle with that area, Ra.
> 
>  Then you do the same with the perimeter: calculate it and then
>  calculate the correspondent radius of a circle whit that perimeter,
> Rp.
> 
>  Finally you calculate Ra/Rp (so you get a no dimensional number,
> which is always a good thing
>  when comparing geometrical features, and you don't want to calculate
>  Rp/Ra in order to avoid posible divisions by zero), and the closer
> this value
>  gets to 1, the "more circular" is the figure.
> 
>  I guess this value is 1 if and only if the figure is a circle, for
> the rest of the figures it's always
>  bigger or smaller (I don't remember which is the actual case).

This sound like what I know as "roundness".
This can be defined as:
     4*Area / (pi*Diameter*Diameter).

As you suggest, this is 1.0 for a circle.

I looked at this for the original poster.  I
rejected it because the measure can be less
than 1 or greater than 1 for a non-circle
depending on how the shape is orientated.

Imagine a square with side L.  Depending on
its orientation, the "diameter" can vary
between L and sqrt(2)*L.  The latter case is
when it is a diamond.

Working this through, we get:
     0.64 < roundness < 1.27

At some orientation, the roundness of the
square must be 1.0, the same as for a circle.

-- 
Regards,
Martin Leese
E-mail: please@see.Web.for.e-mail.INVALID
Web: http://members.tripod.com/martin_leese/

Message-ID:
Subject:

Re: object recogniction from edge map

Date:Thu, 8 Jan 2009 21:19:23 +0100


Martin Leese wrote:

> This sound like what I know as "roundness".
> This can be defined as:
>     4*Area / (pi*Diameter*Diameter).

Drat, forgot the reference:
http://ams.allenpress.com/perlserv/?request=get-document&doi=10.1175%2F1520-0469(2003)060%3C1795%3ARAAROP%3E2.0.CO%3B2

-- 
Regards,
Martin Leese
E-mail: please@see.Web.for.e-mail.INVALID
Web: http://members.tripod.com/martin_leese/

Message-ID:<37a9c9fd-5899-4641-a081-5fe3c7d967d1@v18g2000pro.googlegroups.com>
Subject:

Re: object recogniction from edge map

Date:Fri, 9 Jan 2009 11:21:20 +0100


On Jan 8, 9:17=A0pm, Martin Leese 
wrote:
> This sound like what I know as "roundness".
> This can be defined as:
> =A0 =A0 =A04*Area / (pi*Diameter*Diameter).
>
> As you suggest, this is 1.0 for a circle.
>
> I looked at this for the original poster. =A0I
> rejected it because the measure can be less
> than 1 or greater than 1 for a non-circle
> depending on how the shape is orientated.

That depends on how you choose to define the
"diameter" of a shape that is not a circle. For
example, you can define it as twice the average
euclidean distance from perimeter points to the
centroid, and that is rotation-invariant.

Another way to measure roundness could be:
    roundness =3D Rmin / Rmax
where Rmin is the *minimum* Euclidean distance
from a perimeter point to the centroid, and Rmax
is the maximum such distance.
This measurement yields a value of 1.0 (or marginally
below that, due to finite pixel size) for a circle, and
something between 0 and 1 for other shapes.

Message-ID:
Subject:

Re: object recogniction from edge map

Date:Fri, 9 Jan 2009 15:11:00 +0100

On Jan 5, 1:57=A0pm, SUMAN  wrote:
> Hi,
> I am doing machine vision project in Spartan3a dsp 1800a board. I have
> completed sobel edge detection in hardware . Now I have boundary of
> the single object. My problem is to identify whether the given
> boundary is of cube or sphere. I am planning to complete this portion
> using microblaze processor . Can any body suggest me simplest
> algorithm or methods for this.
> Thank you

What are the original images like? Are they photographs of a
particular ball and a particular cube? Is the distance to the camera
fixed, the lighting? What variation is expected? What is the
background like? Or are they synthetic binary images? The fact that no
one has asked you these questions should worry you.

illywhacker;

Message-ID:<7007680f-74bf-437b-99c6-44002f50e662@z6g2000pre.googlegroups.com>
Subject:

Re: object recogniction from edge map

Date:Sat, 28 Feb 2009 09:09:27 +0100


The images are taken from c3038 sensor module. My problem is to
identify cube from the sphere with camera height fixed. I have edge of
image in binary format ant there are not any other background objects.
suman

On Jan 9, 6:11=A0am, illywhacker <illywac...@gmail.com> wrote:
> On Jan 5, 1:57=A0pm, SUMAN <suman...@gmail.com> wrote:
>
> > Hi,
> > I am doing machine vision project in Spartan3a dsp 1800a board. I have
> > completed sobeledgedetection in hardware . Now I have boundary of
> > the singleobject. My problem is to identify whether the given
> > boundary is of cube or sphere. I am planning to complete this portion
> > using microblaze processor . Can any body suggest me simplest
> > algorithm or methods for this.
> > Thank you
>
> What are the original images like? Are they photographs of a
> particular ball and a particular cube? Is the distance to the camera
> fixed, the lighting? What variation is expected? What is the
> background like? Or are they synthetic binary images? The fact that no
> one has asked you these questions should worry you.
>
> illywhacker;

Message-ID:<7007680f-74bf-437b-99c6-44002f50e662@z6g2000pre.googlegroups.com>
Subject:

Re: object recogniction from edge map

Date:Sat, 28 Feb 2009 09:09:27 +0100


The images are taken from c3038 sensor module. My problem is to
identify cube from the sphere with camera height fixed. I have edge of
image in binary format ant there are not any other background objects.
suman

On Jan 9, 6:11=A0am, illywhacker <illywac...@gmail.com> wrote:
> On Jan 5, 1:57=A0pm, SUMAN <suman...@gmail.com> wrote:
>
> > Hi,
> > I am doing machine vision project in Spartan3a dsp 1800a board. I have
> > completed sobeledgedetection in hardware . Now I have boundary of
> > the singleobject. My problem is to identify whether the given
> > boundary is of cube or sphere. I am planning to complete this portion
> > using microblaze processor . Can any body suggest me simplest
> > algorithm or methods for this.
> > Thank you
>
> What are the original images like? Are they photographs of a
> particular ball and a particular cube? Is the distance to the camera
> fixed, the lighting? What variation is expected? What is the
> background like? Or are they synthetic binary images? The fact that no
> one has asked you these questions should worry you.
>
> illywhacker;