detecting rectangle in real images

Hey all,

I want to detect and track rectangle pattern . these pattern I want to set on object which moving and captured using webcam.

So, the image will include real-world scene also - not only the pattern.

example of pattern:

I want to be able to detect the 4 corners of the white rectangle.

what about if I have black rectangle?

Thanks in advance

[Image Analyst]

This does not look like a real world image. It looks like a computer graphics image. Please attach the real world images.

[shdrone]

yes I know- I want to print this pattern and put it on moving object (for example : human) and the camera capture image including the pattern and more.

I wrote this note because I found a code that can detect this rectangle when you pass to it the pattern image only - and detect another objects in the image that isn't a rectangle when you pass to it real world image.

[Image Analyst]

Okay, so....after that explanation . . . my comment remains the same.

[shdrone]

ok - no problem.

here's an image :

thanks

[shdrone]

Do you have a solution for this image?

Answer 1

[Image Analyst]

Does the clutter have to be in the background? It requires a more time consuming and complicated algorithm. I'd take the green channel, threshold for dark things. Find objects in a certain size range with a certain Euler number. Then call bwboundaries() and take the inner boundary.

[shdrone]

Hi,

I found this : http://www.mathworks.com/matlabcentral/answers/uploaded_files/8433/shape_recognition.m

I think it's working good (I tested some images).

my question : can this code give me the coordinates of the 4 corners of the rectangle or just the Centroid?

Thanks for your help

[Image Analyst]

No it can't (and that's my code so I should know). But I have an updated demo, attached, that can.

Answer 2

[Spandan Tiwari]

If your rectangle does not change orientation, you can try using normxcorr2 (in Image Processing Toolbox) with the rectangle as the template. If changes in orientation are possible you may have to use more than one templates.

Answer 3

[Elizabeth Rocha]

my code for this is :

imagen=imread('img18.jpg'); 
recorte=imagen(112:221,218:274); 
recorte2=imagen(261:309,233:317); 
recorte=im2bw(recorte); 
recorte2=im2bw(recorte2); 

subplot(1,3,1),imshow(imagen)
rectangle('position',[218,112,56,109]);

% ---this is for captured using webcam 
rectangle('position',[233,261,84,48]); 
subplot(1,3,2),imshow(recorte) 
subplot(1,3,3),imshow(recorte2) 

if recorte==true 
    x=1 
else
    x=0 
end

if recorte2==true 
    y=1 
else
    y=0 
end

[Ayush singhal]

Can you tell how to generate rectangular stripes with fixed period and width of the stripes should be half of period.

[Image Analyst]

@Ayush singhal, try rectpuls() if you have the Signal Processing Toolbox. Otherwise create one pulse using ones() and zeros() and then replicate it as many times as you want with repmat.

[Ayush singhal]

But I need stripes as a image not like pulses. stripes like black white black white black white.....

I think this needs image processing tool.

Answer 4

[Matt J]

This works for the one given image. I would need more examples to see if it's a reliable solution. It uses the FEX contribution pgonCorners which must be Downloaded.

load Image

C=bwpropfilt(all(Image<50,3),'ConvexArea',1);

D=bwareafilt( bwconvhull(C)-C>0.5 ,1);

xy=fliplr(pgonCorners(D,4)); %corner coordinates

imshow(Image);
hold on
scatter(xy(:,1), xy(:,2),30,'r','filled')
hold off